firecrawl - Research Engineer — Reinforcement Learning

• Someone who builds their own training infra and reward pipelines. You don't wait for an ML platform team to set things up. You build the training loops, reward models, data pipelines, and evaluation frameworks yourself — because you understand that the infra choices directly affect the quality of the results. You've operated GPU clusters, managed training runs, and debugged convergence issues in production. • Can fine-tune models to achieve SOTA results. You've taken models from baseline to best-in-class on tasks that matter. You understand the full fine-tuning lifecycle — data curation, training dynamics, hyperparameter sensitivity, evaluation methodology — and you have the taste to know when a model is actually good versus when the eval is flattering. • Bridges LLM agents and classical RL approaches. You're fluent in both worlds. You understand PPO, RLHF, reward modeling, and policy optimization — and you also understand how modern LLM agents work, where they fail, and how RL techniques can make them better. You see connections between these domains that most people miss. • Runs fast experiments and communicates clearly. You bias toward quick iterations over perfect setups. You'd rather run three rough experiments this week than one polished one next month. And when you have results, you communicate them clearly — to other researchers, to engineers, and to leadership. No one needs to decode your work to understand its impact. • Production-minded. You care about whether your models actually work in production, not just on benchmarks. You've deployed models that serve real traffic and you've made hard tradeoffs between model quality, latency, and cost. Research that doesn't ship isn't research that matters here. • Backgrounds that tend to do well: RL engineers at AI labs or applied ML teams who've shipped models to production. Researchers who've done RLHF or reward modeling for LLM systems. ML engineers who've built training infrastructure at startups and cared as much about the pipeline as the model. People who've worked at the intersection of RL and language models — whether in academic labs with a production bent or at companies building agent systems. • WHAT WE'RE NOT LOOKING FOR • Pure theorists. If your best RL work is a proof in a paper and you've never trained a model on real data at real scale, this isn't the role. We need someone who builds and ships, not someone who derives. • Researchers who need a platform team. If you expect training infrastructure, data pipelines, and evaluation frameworks to be set up for you before you can be productive, you'll be frustrated here. You build the tools you need. • People who only know one paradigm. If you're deep in classical RL but have never worked with LLMs, or if you're an LLM fine-tuner who's never touched RL — you'll be missing half the picture. This role requires fluency in both. • Slow iterators. If your standard experiment cycle is measured in weeks, not days, you'll struggle with the pace here. We need someone who can run a meaningful experiment, interpret results, and decide next steps within a day or two — not someone who needs a month-long study to make a call. • Black-box communicators. If your typical update is a wall of metrics that only another RL researcher can interpret, this isn't the right fit. We need someone who can explain what's working, what's not, and why it matters to people who don't have RL PhDs. • We operate at an absurd level of urgency because the window for what we're building won't stay open forever. If that excites you, keep reading. If it doesn't, no hard feelings — but this role probably isn't for you.