Immunology Scout

Immunology Scout is intentionally scoped as a research grounding layer rather than a broad "AI for science" platform.

The current wedge is paper + patent exploration: combine scientific literature and prior-art signals so teams can see what is known, what is uncertain, and where genuine opportunity may exist.

Patent-aware exploration helps researchers pressure-test novelty, understand competitive context, and avoid duplicating work that may already be claimed or crowded.

Outputs are structured to support hypothesis refinement, competitive analysis, and future downstream workflows such as simulation and experimental planning systems, while keeping human scientific reasoning in the loop.

This was the first project where I built formal evals. Because I was not the domain expert, I worked with a scientific collaborator to define what a useful, trustworthy output should actually contain.

Structured expert feedback became the basis for a lightweight evaluation set: what evidence should be cited, where claims needed tighter boundaries, and what would make the product genuinely useful to a researcher instead of merely polished.

That process surfaced a critical issue early: the system could hallucinate patent references, which is unacceptable in a workflow meant to support novelty checking and competitive understanding.

Once those failures were visible and measurable, iteration became much more disciplined. Later versions enabled tighter automated feedback loops, but the larger lesson was product-oriented: in high-stakes domains, trust comes from repeated loops between domain expertise, evaluation, traceability, and system behavior.