How we build
We ship working software, not vibes. We write Python other humans can read, add tests where they are needed, keep data and models versioned, and make work reproducible on a fresh laptop. End-to-end ownership is real: clarify the problem, pick the simplest plan, measure outcomes, and iterate without ego. CI/CD for the whole pipeline, telemetry that catches drift before customers do, and governance that isn't theater (validation, experiment QA, explicit assumptions).
Open Positions
Data Scientist
Turn fuzzy questions into measurable wins. You design credible experiments, separate cause from coincidence, forecast what's next, map behavior, spot outliers before they bite, and, when it actually helps, guide choices with ranking or recommendations.
Your SQL is crisp, your models are explainable and calibrated, and your metrics tell the truth. You make decisions easy for busy people: clear options, real trade-offs, no drama. 3+ years in the wild helps, but sharp thinking beats shiny titles.
AI Engineer
Translate model capacity into reliable, efficient behavior. You understand how tokens, attention, and embeddings drive capability, and you improve models with practical tuning strategies, curated data, reproducible experiments, and steady iteration (not “just throw a bigger model at it”). When correctness matters, code, math, structured tasks, you design preference / objective-driven training and checks that keep the model honest.
You build evaluations that blend automation with human judgment, track quality like a product metric, and raise the bar for reasoning and alignment. When it's time to serve, you engineer for reality: squeeze latency/throughput/cost with sane batching/caching/quantization, choose the right memory trade-offs, roll out with canaries or A/B, and push to edge when privacy or speed demands it. You think in systems and prototype fast with product partners to get to "shipped".
How to apply
Send us a short message. Subject: "I make models useful". Body: Three lines, (1) what you built, (2) why it worked, (3) proof it was impactful. Attach nothing, we'll ask for more if we're intrigued.