Staff Research Engineer (LLM Pre-Training)

Wizard

ABOUT WIZARD Wizard is the top-performing AI Shopping Agent, delivering the best products from across the web with unmatched accuracy, quality, and trust. THE ROLE We’re looking for an Applied Scientist to own how we measure, understand, and improve the accuracy of our AI agent. This role sits at the intersection of applied ML, evaluation science, and product. You’ll define what “good” looks like for our agent, build the systems to measure it, and lead the science work to improve it, including fine-tuning the LLM judges that power our evaluation pipeline. You’ll partner with ML Engineering and AI Engineering. What you will do is bring scientific rigor to the most important question at Wizard: is our agent getting better, and how do we know? This is a foundational hire on our science team. Evaluation is the starting point, and the role is scoped to grow into broader applied science work as the surface area of the agent expands (recommendations, personalization, ranking, multimodal, conversational understanding). WHAT YOU’LL DO * Define and evolve accuracy metrics across the full shopping experience (retrieval, ranking, recommendations, outcomes) * Design and run experiments to measure improvements and regressions * Build and maintain evaluation datasets, benchmarks, and scoring frameworks * Improve the LLM judges that power our evaluation pipeline: prompting, calibration, and fine-tuning where it matters * Translate ambiguous product questions into clear, measurable hypotheses and analysis * Partner with ML Engineers to validate model changes and guide iteration * Identify failure modes and edge cases, and drive improvements through data * Make agent performance visible, trusted, and actionable across product and engineering FIRST 3 MONTHS * Go deep on the agent, the current eval pipeline, and the metrics we use today * Audit existing accuracy metrics and benchmarks; identify gaps, blind spots, and signals that aren’t trustworthy * Build relationships with ML, AI Engineering, and Product * Ship one quick win: a missing benchmark, an improved metric, or a fix to a misleading signal * Establish a baseline view of agent performance the team can rally around MONTHS 3 TO 6 * Own the evaluation framework: datasets, metrics, scoring, reporting, both offline and online * Drive measurable improvements to LLM judge quality (calibration, fine-tuning where appropriate) * Run experiments that influence at least one significant model or product change * Stand up automated evaluation the team trusts before and after every launch * Build dashboards and reporting that make agent performance legible to leadership BEYOND 6 MONTHS * Lead applied science work on the next frontier as the agent grows: multi-turn evaluation, multimodal, personalization, ranking quality, conversational understanding * Influence team-level strategy on what we measure, what we improve, and why * Mentor and help grow the science function as it expands WHAT SUCCESS LOOKS LIKE * Clear, trusted accuracy metrics are consistently used across product and engineering * A robust automated evaluation framework for both offline and live experiments * Model and product changes are consistently measured before and after launch * Demonstrable improvements in LLM judge quality and eval coverage * Science leadership that informs what we build, not just whether it works CAREER GROWTH * Depth track: become the org’s authority on AI evaluation: eval strategy, judge models, agent benchmarking * Breadth track: expand into other applied science problems (recommendations, personalization, ranking, multimodal, conversational understanding) as those areas come online * Leadership track: Senior / Staff Applied Scientist, with technical leadership across the science function * As the agent gets more capable, the science problems get richer IDEAL BACKGROUND * 5+ years in Applied ML, AI Research, or Applied Science (PhD or equivalent depth strongly preferred) * Hands-on experience evaluating modern AI/ML systems: LLMs, agents, ranking, or recommendations * Direct experience with LLM-based systems: judge models, RAG, prompt engineering, fine-tuning, RLHF, or similar * Strong experimentation foundations: A/B testing, causal inference, statistical rigor * Proven ability to operate in ambiguity: defining problems, not just solving pre-defined ones * Clear, structured communication that influences across ML, engineering, and product COMPENSATION & BENEFITS The expected base salary range for this role is $225,000 - $280,000 USD, and will vary based on skills, experience, role level, and geographic location. Final compensation will be determined by considering these factors alongside overall role scope and responsibilities. In addition to base salary, Wizard offers: * Equity in the form of stock options * Medical, dental, and vision coverage * 401(k) plan * Flexible PTO and company holidays * Fully remote work within the United States * Periodic company offsites and team gatherings Wizard is committed to fair, transparent, and competitive compensation practices.

Black Forest Labs

ABOUT BLACK FOREST LABS We're the team behind Latent Diffusion, Stable Diffusion, and FLUX—foundational technologies that changed how the world creates images and video. We’re creating the generative models that power how people make images and video—tools used by millions of creators, developers, and businesses worldwide. Our FLUX models are among the most advanced in the world, and we're just getting started. Headquartered in Freiburg, Germany with a growing presence in San Francisco, we’re scaling fast while staying true to what makes us different: research excellence, open science, and building technology that expands human creativity. WHY THIS ROLE Large-scale training is where research ideas become real, and where many of the hardest problems are no longer cleanly separated into “research” or “engineering.” A promising architecture only matters if we can train it stably, efficiently, and correctly across large GPU fleets. In this role, you will be embedded in production training and help where the hardest systems and performance problems arise: attention performance, custom kernels, low-precision training, profiling, memory behavior, data movement, distributed training stability, and throughput regressions. You will work directly with researchers, but your output will often be code, measurements, kernels, debugging tools, and training-system changes that make better research possible. We are open to a range of seniority for this role. The common thread is deep technical ownership: you should be able to make progress in ambiguous training-system problems, verify your results, and own the outcome. WHAT YOU’LL WORK ON * Improve the performance, reliability, and numerical stability of production training runs for large multimodal generative models * Profile full training steps across model code, attention, kernels, data loading, encoders, communication, optimizer steps, checkpointing, and memory pressure * Implement and validate GPU-level optimizations: fused kernels, attention paths, low-precision matmuls, quantization kernels, CUDA/Triton/CuTe/CUTLASS experiments, and no-compile alternatives where they make sense * Push lower-precision training forward, including FP8 / MXFP8 / FP4-style paths, weight and activation quantization, accumulation choices, convergence risk, and quality tradeoffs against baseline training runs * Work with researchers to translate architecture changes into efficient training implementations, and help distinguish real model-quality progress from changes that only look good in a microbenchmark * Debug distributed training failures: NaNs, loss spikes, silent numerical drift, memory leaks, stragglers, bad nodes, NCCL issues, and throughput cliffs * Build benchmarking and profiling harnesses that make performance claims trustworthy across hardware, shapes, sequence lengths, and training configurations * Help the training team move quickly when an urgent bottleneck appears, while turning repeated failures into better abstractions and tools WHAT WE’RE LOOKING FOR * Experience working deeply on large-scale training systems, ideally as part of a training group working closely with researchers * Strong PyTorch fluency, including comfort reading and modifying low-level training code rather than only using high-level APIs * Experience with distributed training concepts such as FSDP, tensor/model/context/sequence parallelism, activation checkpointing, NCCL, and overlapping compute and communication * Hands-on experience improving training throughput, memory footprint, or stability in real training runs * Experience profiling GPU workloads with tools like Nsight Systems, Nsight Compute, torch profiler, trace viewers, or custom telemetry * Practical GPU performance judgment: you may use modern coding agents and tools as much as you want, but you need the understanding to verify correctness, numerical behavior, and performance, and to own the result * Understanding of low-precision training and quantization tradeoffs: FP8, MXFP8, FP4/NVFP4-style formats, scaling, accumulation, numerical validation, and convergence risk * Good research judgment: you can partner with researchers on ablations, understand what the measurements do and do not prove, and keep optimization work tied to model-quality outcomes * Comfortable operating in ambiguity: sometimes the task is a clean implementation, sometimes it is a production fire, and sometimes it is figuring out which of three plausible explanations is actually true We'd be especially excited if you: * Have supported or co-owned training for a frontier foundation model that shipped or reached a major release * Have written or substantially improved forward/backward GPU kernels, or have shown you can make progress on kernel-level work with strong measurement and validation discipline * Have worked on attention performance, variable sequence length training, non-standard attention patterns * Have experience on Hopper or Blackwell-class GPUs * Have worked on low-precision training * Have experience with diffusion, flow matching, DiT, and multimodal generative model training; if your deepest background is autoregressive or LLM training systems, you are excited to learn the diffusion and multimodal modeling stack quickly * Can move naturally between profiler traces, kernel code, distributed systems failures, and research discussions HOW WE WORK TOGETHER We’re a distributed team with real offices that people actually use. Depending on your role, you’ll either join us in Freiburg or SF at least 2 days a week (or one full week every other week), or work remotely with a monthly in-person week to stay connected. We’ll cover reasonable travel costs to make this possible. We think in-person time matters, and we’ve structured things to make it accessible to all. We’ll discuss what this will look like for the role during our interview process. Everything we do is grounded in four values: * Obsessed. We are a frontier research lab. The science has to be right, the understanding deep, the product beautiful. * Low Ego. The work speaks. The best idea wins, no matter who said it. Credit is shared. Nobody is above any task. * Bold. We take the ambitious bet. We ship, we do not wait for conditions to be perfect. * Kind. People over politics. We treat each other with genuine warmth. Agency without empathy creates chaos. If this sounds like work you’d enjoy, we’d love to hear from you. Base Annual Salary: EU €130,000 - €240,000 + Equity US $180,000 - $290,000 + equity

Unlikely AI

STAFF SOFTWARE ENGINEER At UnlikelyAI, we are building the future of AI: one that is reliable, accurate, and transparent. Our neurosymbolic technology harnesses the power of LLMs and generative AI, and combines it with Universal Language – our proprietary symbolic technology that bridges the gap between probabilistic machine learning and deterministic classical computing. Our products are already in use with major enterprises – including tier-1 banks and leading accountancy firms – across audit, compliance, and financial services. In compliance, we combine symbolic decision trees with LLM-powered evidence extraction to catch errors in financial reporting that human reviewers miss. In financial services, we use neurosymbolic guardrails to deliver accurate and explainable outcomes at scale. We are now building toward a platform – a public API and platform experience that will make our core neurosymbolic capabilities available to a broader set of customers and use cases. This is a pivotal moment: we're transitioning from bespoke customer engagements into a scalable product platform, and we need exceptional engineers to help us get there. THE ROLE We are looking for a Staff Software Engineer to help shape the technical direction of our platform as we scale. This is a role for someone who combines deep hands-on engineering ability with the judgement and influence to drive architecture and engineering quality across teams. You'll be one of our most experienced individual contributors – someone the team looks to for guidance on hard technical decisions, system design, and long-term technical strategy. You'll spend most of your time writing code and solving complex problems, but you'll also be expected to identify the highest-leverage work across squads, mentor other engineers, and raise the bar for how we build software. Our core capabilities span symbolic reasoning (decision trees, propositional graphs, knowledge graphs), document ingestion pipelines, and the APIs that expose these to customers. You'll work on genuinely novel problems at the intersection of classical symbolic AI and modern LLMs – for example, how to represent regulatory knowledge as machine-evaluable rules, or how to build feedback loops that improve system accuracy over time. You'll work within a shared monorepo alongside software engineers, research engineers, and applied scientists in a heavily cross-functional environment. We operate in small, focused product teams, supported by shared infrastructure, internal tooling, and an R&D function. WHAT YOU MIGHT WORK ON In your first months, you could find yourself working on any of the following: * Defining the architecture for our new public API – making foundational decisions about authentication, scalability, versioning, and developer experience that will shape the platform for years. * Leading the design and implementation of our document ingestion pipelines to handle new input formats (e.g. PDF, Word) and new regulatory jurisdictions at scale. * Designing evaluation frameworks and benchmarks to measure and improve system accuracy – and establishing these as engineering norms across teams. * Driving improvements to our deployment architecture for enterprise customers with specific cloud and security requirements. * Owning the technical strategy for internal tooling and developer experience across the monorepo – identifying bottlenecks and leading initiatives to address them. * Working on the symbolic reasoning engine that powers our products – including decision tree evaluation, rule generation, and knowledge graph construction. * Identifying and leading cross-cutting technical initiatives that improve reliability, performance, or engineering velocity across the organisation. YOU'LL BE SUCCESSFUL HERE IF... * ...you have deep expertise in Python, including writing well-typed, well-tested code in a collaborative codebase, and strong opinions on how to structure Python projects at scale. * ...you have a proven track record in system design and architecture – you've made foundational technical decisions that shaped the trajectory of a product or platform. * ...you've tackled complex algorithms and data structures and have experience working with non-trivial algorithmic problems at scale. * ...you care deeply about production-quality engineering – you don't just advocate for software quality, you actively set the standards and build the culture around it. * ...you have a track record of technical leadership – you've influenced technical direction across multiple teams or projects without necessarily having direct reports. * ...you have significant experience with cloud infrastructure (AWS preferred) – services such as S3, ECR, ECS/EKS, and infrastructure managed via Terraform or similar – and can make informed architectural decisions about deployment and scalability. * ...you have a bias for action – you move quickly, make informed decisions, and iterate without waiting for perfect information. * ...you have a relevant degree in Computer Science, Mathematics, Engineering, or STEM – or equivalent practical experience. OTHER SKILLS You don't need to tick every box below, but any of the following would strengthen your application: * Monorepo experience – comfortable working in and improving a large, shared codebase with multiple product teams contributing. * CI/CD pipelines – hands-on experience with GitHub Actions or similar, ideally including designing and optimising CI infrastructure. * Experience with document processing pipelines – PDF parsing, OCR, structured data extraction. * Familiarity with knowledge representation – decision trees, knowledge graphs, ontologies, or symbolic reasoning systems. * Experience with LLM integration in production systems – prompt engineering, evaluation, working with APIs such as Gemini, Claude, or OpenAI. * Frontend experience with React and TypeScript – we value engineers who can contribute across the stack when needed. * Experience in regulated industries – fintech, audit, compliance, insurance, or banking. * Familiarity with the modern Python tooling ecosystem: uv for package management, ruff for linting, pyright or similar type checkers. * Experience with observability and monitoring tools such as Datadog. * Experience mentoring engineers and helping teams grow their technical capabilities. HOW WE WORK We're a team of around 30 people based primarily in the UK. We operate a hybrid working policy, with three days a week in our Central London office. Engineering is organised into product-focused squads, supported by shared infrastructure and an R&D function. We work in a monorepo, deploy to AWS, and care deeply about developer experience – we're actively investing in modernising our tooling, CI, and repository structure. We run hackathons, we have strong opinions about code quality (held loosely), and we ship often. Our culture is collaborative and low-ego: engineers regularly move between teams, pair on hard problems, and contribute ideas regardless of seniority. We take the work seriously, but not ourselves.