# Poolside Interview Guide (2026): How to Prep for a Code-Model Lab Source: https://www.techinterview.org/companies/poolside-interview-guide/ Updated: 2026-07-21 · techinterview.org **TL;DR —** Poolside interviews skip the LeetCode gauntlet and test whether you can reason about training and running a code-generation foundation model. Expect a short loop, roughly a week, that opens with a founding engineer, runs through two technical rounds tailored to your track, and closes with a culture and founder conversation. Research-track candidates get pushed on pre-training, RL from code execution feedback, evals, and scaling; engineering-track candidates get pushed on distributed systems, data pipelines, and inference. You stand out with specific opinions backed by experiments you actually ran, not memorized answers. Poolside builds foundation models aimed squarely at software development. It was co-founded in 2023 by Jason Warner, GitHub's former CTO, and Eiso Kant, who earlier started the developer-tooling companies source{d} and Athenian. The company's public bet is RLCEF, reinforcement learning from code execution feedback: instead of only predicting the next token, its models train against signals from actually running code, such as whether it compiles and passes tests. Its lineup has included Malibu for reasoning and chat and Point for fast completion, packaged into a developer assistant and an enterprise platform for fine-tuning and on-prem deployment. Funding has been heavy for the stage: a 2023 seed, a roughly $500M Series B in 2024, and a large 2025 round reported to involve Nvidia that pushed its valuation into the billions, alongside a CoreWeave compute partnership. It runs remote-first, with a US base and a sizable European presence including Paris, and a flat team of around 200. ## Process Poolside runs a short loop, often about a week end to end, and it splits by track. A founding engineer or team lead does the first screen, usually a conversation about your background and why the code-model problem interests you. Two technical rounds follow with Members of Engineering, and candidates report they are tailored to the role rather than pulled from a generic bank: the first digs into the domain you would own, the second plays more like a two-way team overview. A culture-fit call with the People team and a final conversation with a founding engineer close it out. The research track (pre-training, post-training, RL, synthetic data, evals) leans on your ability to design experiments and reason about model behavior; the engineering track leans harder on distributed systems, data infrastructure, and inference. Coding still shows up, but as practical problem-solving rather than a timed [coding](/algorithm-patterns-cheat-sheet/) puzzle, and some rounds can look like [ML system design](/category/system-design/) where you sketch a training or serving pipeline. Expect a [behavioral](/post/3233460379/behavioral-interview-questions-2026-star-method-amazon-leadership-principles-and-winning-answers/) thread throughout, since a flat, remote-first team screens hard for people who own ambiguous work. Reports on the experience are mixed, and a few candidates found scheduling loose, so confirm timelines with your recruiter. ## What they actually ask - How would you design a synthetic data pipeline that produces high-quality, diverse training examples from open-source repositories without blowing up compute? - Walk through reinforcement learning from code execution feedback: what reward signal do you use, and how do you stop the model from gaming the tests? - How would you build an eval suite for a code model that catches regressions across languages and does not overfit to a benchmark like HumanEval or SWE-bench? - Design the distributed training setup for a large model on a big GPU cluster: parallelism strategy, failure handling, checkpointing, and throughput. - How do you serve a code-completion model at low latency and high throughput, and where do you trade off context length, batching, and cost? - A practical Python coding or debugging task tied to data processing or model tooling, talked through rather than graded on raw speed. ## Levels and comp (2026) Poolside does not publish a public leveling ladder, and it uses a flat Member of Engineering style title rather than L3/L4/L5 bands, so comp is negotiated per person against the frontier-lab market. Expect an offer shaped as a strong cash base plus meaningful equity, with the equity being the part that matters most and the hardest to value. Public data points are thin: levels.fyi shows a small number of self-reported packages, with a US software-engineer base reported in the low-to-mid $200Ks and total comp climbing well above that once equity is added for senior research and infrastructure hires. Because the company is private and recently revalued, the equity is a bet on a specific late-stage number, so ask about strike price, the current preferred valuation, vesting, and refresh policy. Treat any single figure with caution and check levels.fyi and recent Blind threads for the freshest ranges before you negotiate. ## Prep priorities Read the company's own writing on RLCEF and be ready to argue its strengths and limits, because you will be asked to reason about it, not recite it. Pick one part of the model lifecycle you know well (pre-training data, post-training, RL, evals, or inference) and go deep enough to defend design choices under pressure. Bring a real project you can dissect: what you tried, what failed, and what the ablations showed. Refresh distributed training basics such as data and model parallelism, mixed precision, and checkpointing, plus modern serving details like KV cache, batching, and quantization. Keep your Python sharp for a clean practical problem, but do not grind puzzles; the bar here is readable code and clear reasoning. Skim recent papers on code generation, RL from execution feedback, and eval design so you can talk shop. Have a few stories ready for a small, high-autonomy team that expects you to move ambiguous problems forward on your own. ## Frequently Asked Questions **Is the Poolside interview LeetCode-heavy?** Candidates consistently report no LeetCode gauntlet. Coding appears as practical problem-solving and debugging, and the technical rounds center on your domain and how you reason about models and systems. **How long does the process take?** Reported loops run about a week, from the first founding-engineer screen through two technical rounds, a culture call, and a final founder conversation, though a private company's scheduling can stretch that. **Do I need published ML research to get hired?** For research-track roles it helps a lot, and job posts list publications as a plus, but hands-on experience with large-scale training, data, or evals can carry an application. Engineering-track roles weight systems and infrastructure over papers. **What makes a candidate stand out?** Specific, well-argued opinions backed by experiments you actually ran, plus the ability to reason out loud about trade-offs in training, evals, and serving. A flat team hires people who can own ambiguous problems with little direction.