Job description

About us

SenseLab is an early-stage startup building the continual-learning engine for AI agents. Agents run and produce outcomes, but most of that signal is lost today. We capture what agents learn from their own runs and turn it into models that improve with use. Backed by a16z Speedrun.

The role

You own retrieval and learning quality end-to-end, from research question to production. This is a founding research seat with a lot of greenfield: you set how we measure quality, decide what earns a place in the serving path, and build the evaluation that keeps it honest. You also publish and speak, because we build in the open.

What you'll work on

Retrieval and ranking. Push relevance on hard queries and design the ranking models that decide what an agent sees.
Evaluation infrastructure. Build the benchmarks and CI gates that catch regressions before they ship.
Continual learning. Own the evaluation science behind our fine-tuning loop: generalization, holdouts, and the bar a tuned model clears before it serves traffic.
Research and publishing. Turn results into papers, blog posts, and talks that bring engineers to us.
Customers. Work with customers' engineers to design, onboard, and implement continual learning for their agents.

Requirements

Shipped retrieval or ranking in a real product.
Strong experimental method: you build the gold set, run the test, and read results honestly.
Model training experience; fluent in PyTorch.
Strong writer who enjoys explaining work in public.
Comfortable in Python and close to the data layer (Postgres, pgvector).

Bonus

PhD in a relevant field (IR, ML, NLP, or similar).
Publications at NeurIPS, ICML, ACL, or equivalent.
Vertex AI or hosted fine-tuning; vLLM or similar serving.
Prior work on retrieval or agent systems.

Not this role

Training foundation models from scratch
Pure backend or infra.
Research that avoids shipping.