Hi Product Hunt 👋 I'm a solo builder with a genetics obsession and an engineering background, and Contig is my attempt at the unglamorous, unsolved part of bioinformatics. Turning "find differentially expressed genes" into a workflow is the easy, crowded part — general LLMs and tools like Galaxy already do it. The hard part is everything after: actually RUNNING that pipeline on real data, on real compute, watching it break at hour 3, and fixing it. Frontier models score only ~17% on real bioinformatics analysis (BixBench), and ~74% of wet-lab scientists don't code at all. So Contig doesn't just generate a workflow. It: • runs the pipeline on YOUR compute (Docker locally, or AWS Batch in your account), • self-heals failures on its own — OOM, a missing index it can rebuild, version mismatches — and shows you the repair chain (risky fixes pause for approval), • verifies the output with real QC checks and ends in an honest verdict (PASS / WARN / FAIL / UNVERIFIED — never a bare "done"), • and writes a signed, reproducible record you can re-run with one command. Being honest about maturity, because that's the whole point of the product: it's Apache-2.0 and free to self-host today. The engine (run → self-heal → verify → reproduce) is exercised end-to-end on nf-core/rnaseq; six more assays (scRNA-seq, germline + somatic variants, methylation, 16S, metagenomics) are wired with QC packs and being hardened to the same bar. The dashboard is localhost-only for now, and it's pre-revenue. Try it in 30 seconds (no data of your own needed): pipx install contig contig run --run-id smoke # runs, self-heals an injected failure, verifies contig show smoke # verdict + provenance + repair chain What I'd love from you: break it, and send me your worst failure cases — every recovered failure becomes a labeled data point that makes the detector better. Happy to answer anything here all day.

Contig

Agentic bioinformatics: it runs, self-heals, and verifies

Agentic bioinformatics: it runs, self-heals, and verifies