What you’ll work on
- Own the evaluations that define a good review: datasets from public, synthetic or explicitly authorised examples, never customer content, scored on precision, recall, latency and cost.
- Improve agents wherever the evaluations point: repository context, review and verification workflows, prompts and tools.
- Post-train models with fine-tuning and reinforcement learning, and ship one only when evaluations show it beats the current setup.
- Harden agents that run on untrusted repositories against prompt injection and tool misuse.
- Carry what works beyond review, to fixes and the routine work that keeps software healthy.
What you’ll bring
- High agency: you find where quality falls short, decide how to fix it and ship the fix without waiting to be assigned.
- A track record of improving models or agents against evaluations rather than demos. Post-training experience, such as supervised fine-tuning, preference optimisation or reinforcement learning, is a strong plus.
- Strong engineering fundamentals: you read unfamiliar code quickly, reason about bugs and work comfortably in TypeScript or Python.