Current AI opportunities

Browse through the roles we're currently hiring for.

Full-time

Applied Researcher, Post-training

San Francisco, California, United States · Remote

USD 200,000-340,000

View role

→

When speech models run in production for thousands of businesses, the hard problems show up in the long tail: a new accent or domain, a latency budget that won’t move, a voice that sounds slightly off for a specific use case. This team exists to solve those problems properly, with research rather than patches. They’re a well-funded speech AI company that trains its own recognition and generation models, and they’re setting up a new applied research group. It’s small, early and close to customers, with a direct line into production. You’d take a problem a customer is seeing, work out what’s actually causing it, and fix it at the model level. Then you’d build the tooling so the next ten problems like it get handled automatically. The work covers: Fine-tuning and adapting ASR and TTS models for specific domains and customers Building evaluation that reflects how models behave in production, not just on benchmarks Data curation and synthetic data generation for targeted improvements Automating the loop from reported issue to deployed model At Staff level, shaping the group’s research priorities across customers You’ll have trained and shipped models that real users depend on, and you’ll be able to talk about how you measured whether they worked. Strong coding, comfort with ambiguity and good judgement about data quality, latency and cost all matter. Speech or audio experience is a plus, but strong applied LLM or fine-tuning backgrounds are relevant too. Being early in a new group means real influence over how it works, and you’ll see your models used at scale quickly. 📍 SF or NYC hybrid, or remote (US) 💰 Up to $340K base + bonus + stock If it sounds relevant, we’d be happy to talk it through. All applicants will receive a response.

Full-time

Member of Technical Staff, Evals

San Francisco, California, United States · On-site

USD 200,000 - 300,000

View role

→

AI Agents | LLM Evaluation | San Francisco How do you know when an AI agent is genuinely ready for production? Not just performing well on benchmarks, but consistently completing real tasks, retrieving the right information and avoiding mistakes that could cost customers money. This is an opportunity to take ownership of that challenge, building the evaluation infrastructure that determines whether AI agents are reliable enough to deploy. You'll join an early-stage AI company developing autonomous agents for logistics operations, tackling complex back-office workflows including appointment scheduling, driver support and load management. Their mission is to build hallucination-free AI for environments where accuracy matters. They're developing agents that operate across text, email and voice, with the longer-term ambition of automating hundreds of operational workflows. The opportunity As a Member of Technical Staff focused on Evals, you'll define how agent quality is measured, tested and continuously improved. This goes well beyond running standard LLM benchmarks. You'll build the evaluation infrastructure used across the company, from engineers developing agents to researchers training proprietary models, and eventually customers validating their own workflows. The challenge is translating complex, real-world interactions into measurable outcomes that accurately reflect whether an agent can do its job. An agent might complete a task successfully once, but can it do so consistently? Is its response grounded in the right information? What happens when conversations become unpredictable or tools return unexpected results? You'll be responsible for answering those questions with reliable evaluation systems. Your technical focus: Build evaluation frameworks measuring task accuracy, pass^k consistency and retrieval quality across individual agents and end-to-end workflows. Develop golden datasets, expert-validated graders and LLM-as-judge systems, including calibration, inter-rater agreement and error severity classification. Implement automated regression testing in CI, with trace-level analysis identifying failures across conversations, tool calls and retrieval. Create simulated users and adversarial test environments for multi-turn text and voice agents, including difficult edge cases drawn from real interactions. A particularly interesting part of this role is connecting offline evaluations with what actually happens in production. You'll analyse production traces, monitor drift and determine whether improvements in evaluation metrics translate into better task completion and automation rates. You'll also help develop self-service evaluation tools, allowing internal teams and customers to create test cases, review failures and validate agents without relying on an engineer for every change. Working alongside the research team, you'll evaluate proprietary model checkpoints and fine-tunes against frontier models, helping determine where improvements are meaningful and where further work is needed. What you'll bring You'll have experience deploying LLM-powered products into production and, importantly, building the evaluation systems that determine whether those products are ready to ship. You'll understand how to turn messy real-world conversations and documents into reliable ground truth, particularly when there isn't a single correct answer. Experience developing LLM-as-judge frameworks, evaluating agentic workflows and validating automated graders against human assessments would be particularly relevant. Strong Python engineering skills, applied statistics and experience integrating evaluations into CI pipelines would also be valuable. Perhaps you've built simulated user environments, adversarial testing systems or evaluation infrastructure for conversational AI products. What matters most is that you've taken responsibility for model or agent quality in a production environment, not simply reported benchmark scores. Why consider it? You'll own a core technical capability rather than contribute to an isolated evaluation project. The infrastructure you build will influence every agent release, how the company measures reliability and how its own models are developed. It's an opportunity to establish evaluation standards from the ground up, with direct responsibility for the systems that determine what ships. Package and location Base salary: $200,000–$300,000 Equity: Meaningful equity package Location: San Francisco, fully onsite If you've built production LLM systems and want to own the evaluation infrastructure behind reliable AI agents, we'd be happy to discuss the opportunity. Contact Techire AI for a confidential conversation. All applicants will receive a response.

Full-time

MTS - Voice Agent

San Francisco, California, United States · On-site

USD 200,000 - 320,000

View role

→

Interested in building the voice agent platform behind AI systems that need to work reliably in the real world? This well-funded AI startup is developing specialist agents that automate complex operational workflows, starting with logistics. Rather than simple voice assistants, they're building agents that work together to manage long-running tasks, from scheduling appointments and coordinating breakdowns to supporting drivers and back-office teams. They're looking for an experienced Voice AI Engineer to take ownership of their voice platform from the ground up, scaling it from early production deployments into a system capable of supporting hundreds of enterprise customers. The technical challenges include: Streaming ASR, TTS and real-time LLM orchestration. Building reliable multilingual, multi-turn conversations with memory and context. Voice agent evaluation, monitoring and production reliability. Developing the infrastructure to deploy and scale voice agents across enterprise customers. Improving latency, conversational quality and agent performance in real-world environments. This isn't about adding a voice interface to an existing product. You'll own the underlying platform, making the technical decisions that determine how voice agents are built, evaluated and deployed at scale. They're particularly interested in engineers who've already built and shipped production voice agents, with experience across speech recognition, conversational AI, multilingual systems or voice evaluations. The founding team has multiple successful exits, including one founder with four previous exits. The engineering team includes several former CTOs, and the company already has paying customers, strong commercial traction and fresh funding, with a larger round planned as they scale. Location: San Francisco, 4 days onsite. Office two blocks from SF Caltrain. Compensation: Up to circa $320K base + stock, healthcare and unlimited PTO. Interested in hearing more about the platform and what they're building? All applicants will receive a response.

Full-time

Inference Systems Engineer

San Francisco, California, United States · On-site

$250,000 - $320,000

View role

→

Inference Systems Engineer Build the distributed inference infrastructure behind safety-focused healthcare AI. Scaling LLM inference isn't just about making individual models faster. It's about how efficiently GPUs communicate, how workloads are distributed across nodes, and how reliably the entire system performs under production traffic. We're working with a well-funded healthcare AI company developing safety-focused foundation models designed to support real-world patient interactions. With over $400M raised and a team bringing together researchers and engineers from Google, Meta, Microsoft, NVIDIA and leading healthcare institutions, they're building AI systems where performance and reliability have direct implications for patient experience. They're now looking for an Inference Systems Engineer to develop the distributed serving and high-performance networking infrastructure behind their LLMs. The opportunity You'll work at the intersection of LLM inference, distributed systems and GPU networking, tackling the performance challenges that emerge when serving large models across multiple nodes. This isn't simply about deploying models through vLLM or SGLang. You'll be working much deeper in the stack, from prefill/decode disaggregation and KV-cache transfer through to GPU interconnects, network topology and communication bottlenecks. Your work will directly influence inference latency, throughput, reliability and cost per token, helping the team deploy increasingly capable models efficiently. Your technical focus Design and optimise disaggregated LLM serving architectures, including independent prefill/decode pools, KV-cache transfer, routing and batching. Improve multi-node GPU communication using technologies such as NVLink, NVSwitch, InfiniBand, RDMA, NCCL and UCX. Build and optimise inference systems using SGLang, vLLM, NVIDIA Dynamo or TensorRT-LLM. Profile performance across compute, memory and networking, improving time to first token, inter-token latency, throughput and tail latency. You'll also develop benchmarking and capacity-planning approaches, troubleshoot complex distributed failures and work with ML researchers to bring new models and hardware into production. What you'll bring You'll need strong experience building high-performance distributed systems, ideally within large-scale ML inference or HPC environments. We're particularly interested in engineers with: Deep understanding of LLM inference, including parallelism strategies, continuous batching, KV-cache management and prefill/decode behaviour. Hands-on experience with GPU networking and communication technologies such as NCCL, NVLink, InfiniBand, RDMA or UCX. Strong Python skills, alongside C++ or another systems programming language. Experience profiling distributed systems and deploying reliable inference workloads on Kubernetes. A Master's or PhD is required. Experience with disaggregated serving, NVIDIA Dynamo/NIXL, CUDA, Triton or low-level GPU optimisation would be particularly relevant. Why consider it? You'll be working on problems that sit right at the intersection of ML systems and high-performance computing, with the opportunity to influence how large-scale inference infrastructure is designed and operated. Rather than focusing on one isolated component, you'll have visibility across the entire serving path, from model execution through to the GPU fabric. And importantly, the systems you build will support healthcare AI applications where low latency, reliability and consistent performance genuinely matter. Package and location Location: Menlo Park, California, five days onsite If you've worked on large-scale distributed inference or GPU networking and enjoy getting into the internals of high-performance systems, this could be worth a conversation. Contact Fraser at Techire AI: fraser@techire.ai All applicants will receive a response.

Full-time

Research Engineer, Benchmarks

San Francisco, California, United States · Remote

USD 100,000 - 230,000

View role

→

Frontier AI Agents | Benchmarking & Evaluations A benchmark is only useful if it actually tells you something meaningful about a model's capabilities. As AI agents become more capable, evaluating them is becoming increasingly difficult. It's one thing to measure performance on a predefined task, but quite another to determine whether an agent can reliably handle complex, real-world workflows. This is an opportunity to tackle that problem, building rigorous evaluation benchmarks that help frontier AI labs understand what their agents can actually do, where they fail and how their capabilities are improving. You'll join a well-funded AI startup building the data infrastructure and evaluation systems supporting the development of frontier AI models. Backed by $16M in funding, including Y Combinator, they're developing a platform used by frontier AI labs, startups and Fortune 500 companies. The opportunity As a Research Engineer, you'll take ownership of developing high-quality benchmarks for evaluating frontier agents across domain-specific tasks. This isn't simply about creating datasets and running existing evaluation frameworks. You'll be investigating what makes a benchmark genuinely useful, from designing realistic tasks and reliable scoring methods to understanding how benchmark results translate into real-world performance. One of the more interesting challenges is developing evaluations for complex workflows where success isn't always straightforward to measure. You'll work with subject-matter experts to understand how tasks are performed in different domains, then translate those workflows into reproducible environments that meaningfully test agent capabilities. The goal is to create benchmarks that frontier AI researchers can trust, rather than another collection of scores that look impressive but reveal very little. Your technical focus: Design and develop agent benchmarks that evaluate realistic, domain-specific workflows, working alongside subject-matter experts to define meaningful tasks. Build infrastructure for reliably running models and agents through benchmark environments, ensuring evaluations are reproducible and consistent. Develop scoring methods and analyses to understand benchmark difficulty, reliability, edge cases and model failure modes. Investigate how benchmark performance correlates with real-world evaluations and produce technical reports that make results credible and interpretable. You'll have significant ownership over the entire process, from initial task design and environment development through to evaluation, analysis and documentation. A particularly important part of the role is understanding why a benchmark works. Can an agent achieve a high score without genuinely completing the intended task? Are there inconsistencies in the evaluation? Does the benchmark discriminate meaningfully between different model capabilities? You'll need to think critically about these questions and build evaluation systems that stand up to scrutiny from experienced AI researchers. What you'll bring You'll have strong software engineering skills, particularly Python, alongside experience working with Docker and Linux environments. Ideally, you'll have worked on agent environments, model evaluations or benchmarking systems and understand the challenges involved in assessing modern AI capabilities. Particularly relevant experience includes: Developing benchmarks, agent environments or evaluation infrastructure. Understanding benchmark design, task realism, scoring reliability and failure analysis. Researching model limitations or evaluating agent performance across complex tasks. Publishing technical research, papers or detailed engineering articles on benchmarks, evaluations or related ML topics. You don't necessarily need years of experience or an extensive publication record. What's more important is your ability to investigate unfamiliar technical problems, understand how real workflows operate and build robust systems that provide meaningful results. Why consider it? You'll join a team of approximately 25 people, including researchers with publications at ICLR and NeurIPS, international Olympiad medallists and experienced AI startup founders. The company is growing, with an opportunity to influence how its benchmarks are designed, evaluated and used by frontier AI teams. You'll have genuine ownership of technically demanding work, with the freedom to investigate problems rather than simply implement predefined solutions. Compensation and location Base salary: San Francisco / major US cities: $135,000–$230,000 Singapore: $100,000–$175,000 Rest of world: $100,000–$175,000 Location: San Francisco or Singapore, with remote options available for candidates who can overlap 70–80% with either office's working hours. Visa sponsorship and relocation support are available for suitable candidates. Benefits include fully covered medical, dental and vision insurance for US and Singapore employees, access to AI development tools, office meals and an annual travel budget. The interview process involves two technical interviews and a 2–3 day work trial. If you've been working on agent evaluations or benchmarking and want to help define how frontier AI capabilities are measured, we'd be happy to discuss the role. Contact Techire AI for a confidential conversation. All applicants will receive a response.

Full-time

Founding Research Engineer, RL/Reasoning

San Francisco, California, United States · On-site

USD 200,000 - 350,000

View role

→

Reinforcement Learning | Healthcare AI | San Francisco Building RL environments for healthcare involves a particularly difficult challenge. How do you design rewards and evaluations that reflect meaningful clinical outcomes, rather than simply optimise what a model can measure? This is an opportunity to tackle that problem, developing the environments and post-training systems that could help AI models reason through complex medical decisions. You'll join a well-funded AI startup building the data and environment infrastructure for medical AI. Backed by Y Combinator and leading venture investors, they're working with AI research labs, healthcare providers and biotech companies to turn complex clinical and biomedical datasets into resources for training and evaluating AI models. Their longer-term ambition is significant: connect clinical histories, genomics, imaging and biological data to help AI systems better understand disease, personalise treatments and accelerate drug discovery. The opportunity As an RL Engineer, you'll work across the full reinforcement learning process, from designing environments and reward functions to training models, evaluating results and improving the underlying data. This isn't simply about applying established RL techniques to another domain. Medical data presents some particularly interesting problems. Patient histories can span years, clinical decisions rarely have simple outcomes, and the quality of training signals can vary enormously. You'll need to understand whether the data genuinely supports the task you're trying to teach a model, and how to build reliable feedback mechanisms around it. That might mean designing environments for clinical reasoning, diagnostic decision-making or chronic disease management, where models learn to navigate complex patient histories, interact with tools and make decisions across multiple steps. Your technical focus: Build RL environments for healthcare, defining tasks, action spaces, tool interfaces, rewards and verifiers. Develop post-training approaches using SFT, RLVR, RLHF/RLAIF and reward modelling. Train and evaluate multi-step agents working with structured and unstructured clinical data. Build scalable rollout, training and evaluation pipelines, using model failures to refine subsequent experiments. A particularly important aspect of this position is data judgement. You'll assess whether datasets have the quality, coverage, label fidelity and longitudinal depth required for meaningful training and evaluation. It's not enough for an environment to produce a reward. That reward needs to represent something useful, and the evaluation needs to tell you whether the model has genuinely improved. You'll have the freedom to investigate different approaches, establish baselines, run controlled experiments and iterate based on what the results actually show. What you'll bring You'll have hands-on experience in reinforcement learning, LLM post-training, agent environments or related ML systems, ideally with at least two years working in these areas. You should understand how to design and evaluate learning environments, build reliable experimental pipelines and debug complex model behaviour. Strong engineering skills are important. You'll be comfortable working in substantial ML codebases, implementing research ideas and investigating why training runs or evaluation systems aren't behaving as expected. Experience with reward modelling, verifiable rewards, agent training or RLHF would be particularly relevant. You'll also need good instincts around data. Understanding what makes a dataset suitable for a particular learning objective is just as important here as knowing which training method to use. You don't need a healthcare background. What matters is your ability to apply rigorous ML research and engineering methods to difficult problems involving real-world data. Why consider it? You'll be working on the infrastructure that determines how medical AI systems learn, rather than building another application around an existing model. There's substantial technical ownership, with the opportunity to define environments, develop evaluation methods and see how your work contributes to practical healthcare AI applications. Package and location Base salary: $200,000–$350,000 Equity: 0.1%–1.0% Location: San Francisco, fully onsite The package also includes medical, dental and vision insurance, 401(k) matching, daily office lunches and an unlimited budget for AI tools and software. Relocation support and joining bonuses may be available, and visa sponsorship is offered for suitable candidates. If applying reinforcement learning to complex medical problems interests you, we'd be happy to tell you more. Contact Techire AI for a confidential conversation. All applicants will receive a response.

Full-time

MTS - Post-Training Lead

San Francisco, California, United States · On-site

$250,000 - $400,000

View role

→

Interested in leading post-training research for multimodal VLMs, working on long-form video understanding, reinforcement learning and spatial reasoning using a large proprietary video dataset? This is a hands-on technical leadership role within a small research team developing models that understand complex real-world environments, with research moving directly into production. You'll work closely with the Chief Science Officer, owning the post-training roadmap across SFT, reinforcement learning, model evaluation and deployment, while leading researchers working on these problems. The team is developing VLMs capable of interpreting detailed activity across thousands of hours of egocentric video, with the longer-term goal of building world models that can predict how physical activities unfold. Your work will include: Leading VLM post-training research using SFT, RL, GRPO and reinforcement learning with verifiable rewards (RLVR). Developing agentic reasoning capabilities that investigate large volumes of video, identify patterns and generate actionable insights. Advancing long-form video understanding, temporal grounding, spatial reasoning and memory across multi-hour footage. Developing models that move beyond activity recognition towards predicting future events and outcomes. Building evaluation frameworks for video understanding, temporal reasoning and real-world model performance. Scaling training and inference pipelines to process thousands of hours of video daily. The team has access to a continuously growing proprietary video dataset, dedicated compute and the opportunity to test new research approaches on complex real-world problems. 💰 Paying up to $400K Base + founding level equity 📍 San Francisco, onsite If you're looking to take ownership of VLM post-training research, lead technical direction and continue developing models hands-on, we'd love to hear from you. All applicants will receive a response.

Full-time

Speech Research Scientist

United States · Remote

USD 200,000-300,000

View role

→

Interested in developing the foundation models behind the next generation of natural, real-time voice AI? An established, well-funded company is growing its research team, building on its own foundation models already powering conversational AI applications at scale. With a strong presence in speech understanding, they're now expanding their research into generative speech, LLMs and multimodal foundation models, with a focus on advancing how AI understands, generates and responds to speech in live conversations. The role As a Senior/Staff Research Scientist, you'll work on some of the key technical challenges in real-time voice AI. This includes LLMs built from scratch and specialised for voice agents, generative speech modelling where naturalness, prosody and control need to hold under strict latency constraints, and perception models capable of handling arbitrary context without compromising responsiveness. What you'll do Build speech foundation models from scratch, spanning understanding and generation Push toward a full speech-to-speech LLM Tackle hard data problems as part of a large-scale expansion of their audio data, training on it as a live stream Own hard problems end to end, from research through to production Work within a small, highly technical team with direct influence on research direction What you'll bring Deep expertise in speech AI, particularly speech generation, Speech LLMs, Speech-to-speech, or speech understanding Hands-on experience developing and training large-scale deep learning models It's a bonus if you have experience with diffusion models or controllable generative modelling. Comp is generally up to $300K cash plus stock and comprehensive benefits, depending on level and location. This is a remote role, however, there are offices within the US if you prefer to be in person.

Full-time

Head of Engineering

San Francisco, California, United States · On-site

USD 200,000 - 325,000

View role

→

Can you take a small team of founding engineers to an engineering organisation built for scale? A well-funded AI company is entering a major growth phase, with an opportunity to build the engineering organisation that will support its next stage of scale. With its technology already in production and growing commercial demand, the company is investing heavily in engineering, expanding its technical capabilities and building the leadership structure needed for long-term growth. They're now hiring a Head of Engineering (Platform) to lead the transition from a small, highly technical engineering team into a mature, scalable engineering organisation. This isn't about inheriting an established structure. You'll have significant ownership over how engineering is organised, how teams operate and how technical decisions are made. Working closely with senior technical leadership, you'll shape both the engineering function and the architecture supporting an increasingly complex production platform. Your focus: Build and scale the engineering organisation, hiring managers, developing technical leaders and introducing clear team structures. Establish engineering processes, technical standards and an operating cadence that supports growth without slowing delivery. Own engineering strategy and architecture across a complex, full-stack production environment. Remain technically involved, reviewing PRs, shaping architectural decisions and contributing code regularly. What you'll bring: Strong engineering background years with significant senior leadership experience. Proven experience scaling engineering teams, managing managers and building organisational structures. Deep hands-on full-stack engineering expertise, ideally across modern JavaScript/TypeScript, backend systems, databases and cloud infrastructure. Experience scaling production systems through substantial growth, ideally within startup or scale-up environments. This is a leadership role, but not one where you'll disappear into meetings. You'll have the autonomy to build the engineering function, develop future leaders and shape technical direction, while staying close enough to the code to make informed decisions. It's an opportunity to combine organisational leadership with genuine technical ownership at a company investing significantly in its next phase of growth. Location: San Francisco, onsite.

Full-time

Staff Research Scientist

San Francisco Bay Area, California, United States · On-site

USD 250,000-350,000

View role

→

Build expressive speech foundation models at trillion-parameter scale, with real world impact. This team has built their entire speech stack in-house, including proprietary LLM-based ASR and TTS, both already outperforming SOTA benchmarks. They're already powering real-time, human-to-AI conversations at scale, every day. Now they're working at trillion-parameter scale to push toward a genuine end-to-end speech-to-speech LLM, one that understands and responds with genuine emotional intelligence and able to have natural, human-like conversations. That means solving problems like long-context reasoning, pronunciation accuracy and maintaining consistency in noisy, real-world environments, in a domain where no model has cracked this yet. They're hiring Senior and Staff-level Speech Scientists (Principal-level also considered) to help drive this frontier work. What you'll do: Build SOTA speech models from the ground up, at genuinely large scale Own problems end-to-end, from research through to production Solve hard, domain-specific speech challenges as part of the push toward speech-to-speech LLM research Help shape technical direction as part of a team working at the frontier of this space What you'll bring: Deep, hands-on expertise in at least one of: Speech/Audio LLMs, TTS or audio generation, speech-to-speech or large-scale speech understanding Experience shipping speech systems at real scale, not just research-stage prototypes Nice to have: Experience with pre-training speech foundation models (HuBERT, Wav2Vec or similar) Experience with Multimodal LLMs especially to reduce hallucinations Experience with Speech LLMs for low latency streaming ASR Package: Up to $250,000-350,000K base (DOE), plus substantial stock. This is an onsite position in South Bay or Seattle.

Full-time

Inference Engineer

San Francisco, California, United States · Hybrid

$200,000 - $300,000

View role

→

Founding Inference Engineer Build the inference infrastructure that makes large-scale LLM deployment faster and more cost-efficient. Most inference platforms are built around GPUs. But what if different stages of LLM inference could run on entirely different hardware, each optimised for the workload? You'll join a seed-funded AI infrastructure startup building a new inference cloud that combines NVIDIA GPUs with specialised AI accelerators to improve performance, latency and cost per token. With significant financial backing and a small founding engineering team, you'll have the opportunity to shape the core architecture rather than simply optimise an existing platform. The opportunity As a Founding Inference Engineer, you'll work at the intersection of distributed systems, GPU computing and LLM serving. The technical challenge is substantial. How do you efficiently distribute inference workloads across different accelerator architectures while maintaining low latency, high throughput and optimal hardware utilisation? You'll help answer that question, designing and building the systems that make heterogeneous inference practical at scale. Your focus: Design and optimise LLM serving infrastructure across GPUs and specialised AI accelerators. Build disaggregated inference systems, including prefill/decode separation and efficient communication between compute resources. Improve continuous batching, KV-cache management, request scheduling and model placement. Optimise inference frameworks such as vLLM, SGLang or TensorRT-LLM, improving throughput, latency and cost per token. You'll work closely with compiler, runtime and infrastructure engineers, profiling bottlenecks across compute, memory and networking. What you'll bring We're particularly interested in engineers who've personally built or significantly improved core inference systems, rather than solely integrated existing platforms. Ideally, you'll have: Strong C++ and Python engineering skills, with hands-on experience building production LLM inference systems. Deep understanding of LLM serving architectures, GPU utilisation and performance optimisation. Experience with vLLM, SGLang, TensorRT-LLM, Triton or similar inference frameworks. Knowledge of distributed systems and the challenges of scaling model execution across hardware. Experience with CUDA, Triton kernels, speculative decoding, tensor parallelism, RDMA or disaggregated inference would be particularly relevant. You don't need experience across every area. Depth in inference engineering and evidence of building core systems matter most. Why join? This is a genuine founding engineering opportunity. You'll influence architectural decisions, work directly on difficult systems problems and help establish the technical foundations before the engineering team scales. You'll have substantial ownership, strong financial backing and the opportunity to build infrastructure that addresses one of the biggest challenges in deploying large AI models: inference efficiency. Package & location Base salary: $250,000–$275,000 (indicative) Equity: Meaningful founding-stage equity Location: San Francisco or New York City If you've been building inference engines, GPU systems or distributed AI infrastructure and want more ownership over the architecture, we'd be interested in speaking with you. Contact Techire AI for a confidential conversation. All applicants will receive a response.

Full-time

Research Scientist

San Francisco, California, United States · Hybrid

USD 300,000 - 450,000

View role

→

Current LLMs are impressive, but they still struggle to reason over long time horizons, maintain context, and operate reliably across complex tasks. This team is tackling those problems head on. They're building a new generation of foundation models designed for persistent state, long-context reasoning, and agentic workflows. Rather than optimising today's architectures, they're exploring what comes next, creating models that can plan, reason, and stay coherent across interactions that stretch far beyond today's context windows. You'll join the research team focused on post-training, reinforcement learning, and model behaviour. Your work will directly influence how these models learn after pre-training, improving reasoning, planning, evaluation, and long-horizon performance while running experiments at a genuinely significant scale. Your work will include: Researching RL techniques for LLM post-training and reasoning Designing large-scale experiments and evaluation frameworks Improving model behaviour across long-context and agent-style tasks Training and evaluating foundation models ranging from 20B to 300B+ parameters We're looking for researchers who have already worked on challenging problems such as RL for language models, post-training, model evaluation, or large-scale experimentation. If you've helped push models beyond benchmark performance and care about how they behave in real-world settings, you'll be in good company. You'll have access to substantial compute, a highly technical research team, and the freedom to explore ambitious ideas that can shape the direction of future foundation models. Salary: Up to circa $450,000 base plus meaningful stock (package negotiable depending on experience) Location: Remote across the US or hybrid in San Francisco. If you'd like to find out more, we'd be happy to have a confidential conversation. All applicants will receive a response.

Full-time

Research Scientist - (RL, Reasoning)

San Francisco, California, United States · On-site

USD 250,000 - 325,000

View role

→

Research Scientist / Senior Research Scientist - Reinforcement Learning, Reasoning Interested in developing AI systems that can reason through complex scientific problems using reinforcement learning and advanced post-training techniques? This is a hands-on research role with a well backed AI startup developing proprietary foundation models and agents for scientific reasoning, trained on large-scale scientific datasets. The research goes beyond optimising models for correct final answers. A key challenge is teaching models to reason effectively through intermediate steps, particularly in scientific domains where ground truth can be expensive, delayed or difficult to establish. Your work will include: Developing reinforcement learning approaches for foundation models, including RLHF, RLAIF and reward-based training. Building process reward models and verifiers that provide fine-grained supervision over intermediate reasoning steps. Researching scalable oversight and alignment methods for complex scientific tasks where reliable ground truth is difficult to obtain. Exploring search, planning and reasoning approaches to improve scientific hypothesis generation and evaluation. Designing new benchmarks and evaluation methods that measure reasoning capabilities on domain-specific scientific problems. Building scalable training pipelines and running large-scale experiments, taking research from initial ideas through to production. At Senior/Staff level, you'll also help define research directions, lead major technical initiatives and mentor other researchers. Experience with reinforcement learning is particularly relevant, whether that's LLM post-training, robotics, game-playing or other RL applications. Knowledge of reward modelling, RLHF/RLAIF, search and planning, or AI alignment would also be valuable. The team is building its own foundation models using proprietary datasets, offering the opportunity to explore fundamental reasoning challenges while seeing your research translated into working AI systems. 💰$275,000 - $325,000 Base + Equity + Lucrative bonus scheme 📍San Francisco If you're interested in advancing reinforcement learning and developing AI systems capable of tackling complex scientific reasoning problems, we'd love to hear from you. All applicants will receive a response.

Full-time

Research Engineer

San Francisco, California, United States · On-site

USD 250,000 - 320,000

View role

→

Interested in solving the scaling challenges behind the next generation of multimodal foundation models? This AI research startup is developing real-time multimodal models that need to train, reason and operate efficiently under production constraints. They're looking for Research Engineers and ML Systems Engineers who enjoy working at the intersection of large-scale model training, distributed systems and inference optimisation. You'll join a small research team of around 15 people, working closely with researchers to build the infrastructure that allows them to train larger models, run experiments faster and deploy their work into production. The technical challenges include: Large-scale distributed training and improving training efficiency. Building training infrastructure and experimentation frameworks. KV caching, inference optimisation and high-performance model serving. Exploring scaling laws and improving model performance. Developing systems that allow researchers to experiment and iterate faster. This is a role for someone who's just as interested in how models are trained and scaled as the models themselves. You'll have significant ownership over the systems behind the research, from training infrastructure through to inference performance, with the opportunity to influence engineering decisions and how the team approaches model scaling. With a relatively small team, there's plenty of scope to take responsibility for difficult technical problems rather than working on one narrow part of a much larger infrastructure organisation. Location: San Francisco, or New York City, onsite (minimum 4 days per week). Compensation: Up to circa $320K base (negotiable) + meaningful stock. Interested in hearing more about the team and what they're building? All applicants will receive a response.

Full-time

Senior Research Scientist

San Francisco Bay Area, California, United States · Remote

USD 200,000-300,000

View role

→

Want to build speech AI that actually sounds human? You'll be joining a well-funded speech AI startup building ultra-realistic voice technology that handles natural laughter, breathing, seamless language switching, and accurate pronunciation across languages and accents. Before training a single model, they built a proprietary fully expressive conversational speech dataset from scratch, grounded in deep linguistic expertise. Now, their architecture separates semantic and acoustic modelling to capture the natural rhythm and timing of real conversation. As their Senior ML Scientist, you'll focus on SOTA speech synthesis modelling and unified S2S architecture. What you'll do Train and evaluate speech synthesis models, iterating on quality and prosodic control Research and develop half and full-duplex multimodal architectures Work across speech representations (neural codecs, semantic tokens, mel features and continuous latents) Build evaluation frameworks that go beyond objective metrics to capture real perceptual quality Work closely with the data team on pipeline quality and curation Work with linguists to ensure modelling and TTS frontend decisions reinforce each other What you'll bring Proven experience with autoregressive and LLM-based audio generation, ideally within TTS Experience with next token prediction using audio codecs Experience working end-to-end - from data, modelling and evaluation Strong PyTorch skills across training loops, distributed training and model internals Nice to have RL and post-training techniques (DPO, SFT) Multimodal next token prediction across audio and text Cross-lingual and multilingual TTS experience Experience with speech-to-speech or full-duplex models such as Qwen Omni, Moshi, etc Experience with codec modelling - EnCodec, DAC, Mimi Hands-on dataset building and pipeline work They've just closed their Series A, generating healthy revenue, and their models are powering hundreds of millions of conversations monthly for major enterprise clients. HQ in SF, but location is flexible. You can be based anywhere, internationally.

Full-time

Research Scientist

San Francisco, California, United States · On-site

USD 250,000 - 400,000

View role

→

Want to build the speech and audio models that define how the next generation of voice AI actually sounds and listens? A well-funded AI startup has developed new model architectures that make real-time conversational AI finally viable at scale. While most voice AI still suffers from delays and computational bottlenecks, they've solved the core efficiency problems that have held the field back. The role As their Senior/Staff Research Scientist, you'll build core speech foundation models that could define the next decade of voice interaction. You'll work on novel architectures that have immediate real-world impact for thousands of customers. What you'll do Design and implement SOTA speech foundation models Develop efficient algorithms for speech processing and audio understanding Create scalable systems that handle massive audio workloads Build comprehensive evaluation methods to validate model performance Collaborate with engineering teams to transition research into production What you'll bring Deep expertise in modern speech technologies (TTS, Speech LLMs, Voice Conversion/Cloning, Speech Translation, ASR, Audio Understanding, Speech-to-Speech) Strong background in generative modelling for audio and speech Publications at leading conferences Track record of implementing research ideas from concept to production You'll join a solid research team, including technical founders who've published work that's fundamentally shifted how the field thinks about efficient, large-scale foundation models. They're well-funded and generating strong revenue. Comp is on par with top AI labs, with base over $400k+ DOE plus a generous equity package. The role is based in SF or NYC (hybrid). If you're excited about building the foundational models that will power the next generation of voice AI, we'd love to hear from you. All applicants will receive a response.

Full-time

Applied Scientist, Post-training

San Francisco Bay Area, California, United States · Hybrid

USD 220,000 - 350,000

View role

→

Interested in developing the post-training methods behind multimodal reasoning models that need to make reliable, verifiable decisions in the real world? This well-funded AI startup is building domain-specific reasoning models that combine text, images and 3D data to plan, retrieve evidence, use tools and solve complex problems in high-stakes industries. They're developing their own foundation models alongside the post-training methods needed to turn them into reliable, agentic systems deployed with enterprise customers. You'll join a growing AI research team, working across model training, reinforcement learning, preference optimisation and evaluation, with the opportunity to take research ideas through to production. The research spans: RLHF, DPO and preference learning , developing post-training methods that improve multi-step reasoning and decision-making. Reward modelling , building reward signals grounded in structured evidence and verifiable outcomes. Multimodal fine-tuning , improving how models reason across text, images and 3D information. Agentic evaluation and RL environments , developing frameworks that stress-test reasoning, tool use and long-horizon behaviour. Training data and experimentation , using proprietary and synthetic datasets to develop and evaluate new training recipes. RL is central to much of the team's research, including the development of their own environments for training, evaluation and testing. They're particularly interested in Research Scientists and Engineers with experience training or post-training transformers, including LLMs and VLMs. Experience with RL, reward modelling, preference optimisation or evaluation frameworks would be especially relevant. You'll have the opportunity to influence how the models are trained, evaluated and deployed, rather than working on isolated research problems. The company has raised $100M and is already partnering with Fortune 100 and 500 customers. Its founder previously built a billion-dollar business, and they're now expanding the AI research organisation towards 40 people. Location: SF Bay Area Compensation: $220K–$350K cash compensation, depending on level and experience, plus equity and benefits. Negotiable for exceptional candidates. Interested in hearing more about the research? All applicants will receive a response.

Full-time

Researcher, RL

London, England, United Kingdom · Hybrid

GBP 120,000 - 220,000

View role

→

Interested in advancing long-horizon reinforcement learning, multimodal agents and multi-step reasoning, with the compute to actually experiment at scale? This well-funded AI startup is developing next-generation Action Models, training large-scale models that can reason, plan, use tools and make decisions across complex, dynamic environments. You'll join a research team working at the intersection of reinforcement learning, reasoning and foundation models, developing new training methods across LLMs, VLMs and alternative architectures such as State Space Models. The research spans: Long-horizon, multimodal and multi-turn RL , training agents to plan and act across extended trajectories. Online and offline RL , improving agent behaviour, alignment and decision-making. Reward modelling and training data , developing better training signals and automated data pipelines. Post-training and reasoning , exploring approaches beyond GRPO to improve model capabilities. One of the biggest attractions is the compute. The team runs its own training, inference and environments entirely in-house, with substantial GPU resources relative to its researcher headcount. You'll have the freedom to run large-scale experiments, iterate quickly and explore new approaches without constantly competing for compute. They're already reporting performance ahead of leading frontier labs in their agent category, while actively publishing research and increasing their public model and research releases. They're interested in Research Scientists and Engineers with experience across reinforcement learning, post-training, reasoning, large-scale model training or multimodal AI. There are also opportunities for engineers specialising in inference and training infrastructure. Location: London (hybrid) Compensation: Up to circa £200K base + equity, negotiable for Staff-level and above. Interested in hearing more about the research? All applicants will receive a response.

Full-time

Research Engineer, Conversational AI

San Francisco, California, United States · On-site

USD 200,000 - 350,000

View role

→

Interested in developing the models behind large-scale conversational AI agents that don't just respond to users, but actually complete complex tasks? This AI company is building the model and decision-making stack behind enterprise agents used by global businesses. Their research spans speech, multimodal understanding and generation, language models and agentic reasoning, with a focus on turning new research into capabilities deployed in production. They're continuing to develop more of their own models, creating opportunities for Research Scientists and Engineers to work across the full model development lifecycle, from training and post-training through to evaluation and deployment. The research spans: Speech and multimodal understanding and generation for natural, real-time conversations. Large-scale model training, instruction tuning and reinforcement learning post-training. Long-horizon reasoning, memory and decision-making for agents completing multi-step tasks. Retrieval, tool use and instruction following within conversational AI systems. Model evaluation and improving performance in real-world enterprise deployments. You'll work closely with researchers and engineers across the organisation, developing models and training methods that improve how agents understand conversations, make decisions and take actions. They're interested in people with experience acrossResearch Scientist / Research Engineer, Multimodal & Conversational AI San Francisco, NYC or London | Up to circa $350K base + equity Interested in developing the models behind large-scale conversational AI agents that don't just respond to users, but actually complete complex tasks? This AI company is building the model and decision-making stack behind enterprise agents used by global businesses. Their research spans speech, multimodal understanding and generation, language models and agentic reasoning, with a focus on turning new research into capabilities deployed in production. They're continuing to develop more of their own models, creating opportunities for Research Scientists and Engineers to work across the full model development lifecycle, from training and post-training through to evaluation and deployment. The research spans: Speech and multimodal understanding and generation for natural, real-time conversations. Large-scale model training, instruction tuning and reinforcement learning post-training. Long-horizon reasoning, memory and decision-making for agents completing multi-step tasks. Retrieval, tool use and instruction following within conversational AI systems. Model evaluation and improving performance in real-world enterprise deployments. You'll work closely with researchers and engineers across the organisation, developing models and training methods that improve how agents understand conversations, make decisions and take actions. They're interested in people with experience across large-scale language models, speech AI, conversational AI, RL post-training or long-horizon agents. You don't need to have worked across both text and voice. Strong experience in either area is relevant. One of the attractions is the scale at which this research is being deployed. The company has raised over $400M, reached a multi-billion-dollar valuation and is approaching nine-figure annual revenue, with established global enterprise customers. You'll be joining an established Research organisation with the opportunity to influence the company's shift towards more proprietary models, while seeing your research make its way into production systems. Location: San Francisco. Onsite working. Compensation: Up to circa $350K base, dependent on level and location, plus equity. Flexibility for senior appointments. Interested in hearing more about the research and the team? All applicants will receive a response.

Full-time

Model Serving Inference

San Francisco, California, United States · Hybrid

$250,000 - £320,000

View role

→

ML Model Serving Engineer Build the inference infrastructure behind the next generation of conversational AI. Making AI conversations feel genuinely natural isn't just a model training problem. It's also about how quickly, efficiently and reliably those models can run in production. We're working with an AI company developing a new generation of voice-based computing, where people can interact with intelligent systems through natural, human-like conversations. Their team brings together founders and engineers with backgrounds at Oculus, Meta, Google and Apple, combining expertise across hardware, ML research and large-scale systems engineering. They're now looking for an ML Model Serving Engineer to optimise the infrastructure serving their LLM, speech and vision models. The opportunity You'll work on some genuinely interesting inference challenges, where milliseconds matter and improvements in throughput, latency and compute efficiency directly influence the user experience. This goes well beyond deploying models using existing frameworks. You'll be modifying and extending frameworks such as vLLM and SGLang, exploring new inference optimisation techniques and working closely with ML researchers to improve model performance without compromising quality. Your work will directly influence how their models are designed, deployed and scaled. Your technical focus Optimise LLM, speech and vision model serving for low latency, high throughput and reliability. Modify and extend vLLM, SGLang and other serving frameworks to improve inference performance. Implement techniques including in-flight batching, caching and custom kernels to reduce inference overhead. Work closely with ML training engineers to identify model-level improvements and reduce model initialisation times. What you'll bring You'll need strong experience optimising ML models in production, combined with a solid understanding of the underlying systems. Ideally, you'll have: Deep PyTorch expertise and experience optimising inference performance at scale. Strong systems programming skills, comfortable working with both complex ML codebases and high-performance serving infrastructure. Experience profiling low-level systems, identifying bottlenecks and improving latency and throughput. Knowledge of LLM serving frameworks such as vLLM or SGLang, ideally including their internals. Experience with Kubernetes, Ray, cloud infrastructure or GPU-level optimisation would also be useful. What's particularly interesting is the relationship between research and infrastructure. You'll have the opportunity to influence both how models are served and how they're developed to perform efficiently in real-world applications. Package and location Salary: $250,000 - $320,000 Location: SF/ NY/ Seattle Benefits include employer-paid medical, dental and vision coverage, 401(k) matching, unlimited PTO and sick leave. If you've spent time digging into the internals of model serving frameworks and enjoy solving difficult performance engineering problems, this is worth exploring. Contact Fraser at Techire AI: fraser@techire.ai All applicants will receive a response.

Full-time

MTS - Research Scientist

San Francisco, California, United States · On-site

USD 175,000 - 300,000

View role

→

Interested in developing multimodal VLMs, working on long-form video understanding and training models on real-world video data that isn't available in traditional research environments? This is a hands-on research role within a small team developing vision-language models capable of understanding complex physical environments, using a large and continuously growing proprietary video dataset. You'll work alongside the Chief Science Officer and other researchers across model training, post-training, evaluation and deployment, with a particular focus on egocentric video understanding, temporal and spatial reasoning, and improving model performance on real-world tasks. Your work will include: Training and improving vision-language and multimodal models using large-scale real-world video datasets. Developing and experimenting with post-training approaches including SFT, reinforcement learning and parameter-efficient fine-tuning. Working on long-context and egocentric video understanding, temporal reasoning and memory. Exploring new training objectives and model architectures for spatial intelligence and fine-grained activity recognition. Building evaluation frameworks for video question answering, temporal reasoning and real-world model performance. Applying distillation, quantisation and model routing techniques to support efficient production deployment. You'll own experiments end-to-end, from data preparation and model training through to benchmarking and deployment, with the opportunity to contribute to the broader research direction as the team grows. Experience with computer vision, multimodal learning, VLMs, video understanding or post-training techniques would be particularly relevant. The team has access to a large proprietary video dataset and dedicated compute, allowing researchers to explore new approaches and deploy model improvements directly into real-world applications. 💰 $175K-$275K base (D.O.E) + significant equity 📍 San Francisco, onsite If you're looking for a research role where you can develop VLMs, explore new training and post-training approaches, and see your research deployed in production, we'd love to hear from you. All applicants will receive a response.

techire AI website footer image