Hi y'all,

I'm relatively new to bioinformatics development and this is one of the first real projects I've built end-to-end. I wanted to share it here because I'd genuinely value feedback from people who work with this kind of data day-to-day - I'm still learning and I know there's a good chance I've gotten some things wrong or missed something obvious, so please don't hold back.

What it does

You search a gene (optionally with a disease context), a variant (rsID), or a disease, and it returns a single source-linked evidence report pulling from:

• PubMed (abstracts + MeSH terms)
• GWAS Catalog
• ClinVar
• Open Targets

Each evidence row links back to its original source. Results are normalized and run through a rule-based scoring layer. I'm intentionally calling this an "evidence triage" signal rather than a "quality" score, since things like risk-of-bias, cohort overlap, or heterogeneity aren't derivable from these APIs - I'd rather be upfront about the limitation than overclaim. There's a /methodology page in the app explaining exactly how retrieval, ranking, and scoring work.

Other things it does:

• Optional grounded LLM summary (the model can only restate retrieved evidence, not add outside claims)
• Gene comparison view (side-by-side score grid for up to 4 genes)
• Citation export (RIS/BibTeX) covering the _full_ literature set behind a report, not just the on-screen sample
• PDF/Markdown report export, saved with a timestamp for reproducibility

Stack: FastAPI/Python backend, React/TypeScript frontend.

Live Demo Link: https://biomedical-evidence-nav.vercel.app/ Repo Link: https://github.com/tunahanf/biomedical-evidence-nav

One heads-up: it's hosted on a free tier, so the backend spins down when idle - the first query after a period of inactivity can take a bit longer (cold start) while it wakes back up. Subsequent searches are fast.

I'm actively improving this and treating it as a learning project, so I'd really appreciate any input, especially on:

• Whether the evidence-triage framing/scoring actually makes sense to people who use this data professionally
• Missing sources or fields you'd expect to see
• Anything that looks wrong, misleading, or naive in how a report is built

Thanks for taking the time to look, and thanks in advance for any critique - that's exactly what I'm hoping to get out of posting this.