GeneTrust AI Studio

Problem

Researchers using CRISPR for gene editing need fast, structured analysis of genetic sequences and predicted edit outcomes. Existing tools either run heavy locally or require deep bioinformatics tooling. The gap: a web platform where biologists submit a sequence and get model-driven analysis with verifiable provenance.

Approach

DNABERT (a BERT pre-trained on the human genome) for sequence understanding, served via a FastAPI inference layer behind a Next.js + TypeScript front-end. Real-time ranking of edit candidates. Each analysis hash anchored on Base Chain for research-grade provenance. A RAG layer cites the published literature behind each prediction.

At a glance

Tech decisions

• DNABERT over training from scratch

  Pre-trained genome embeddings give a credible starting point for downstream classification with limited labeled data.

• FastAPI inference layer

  Async by default; easy per-tenant scaling once research traffic grows.

• Base Chain anchoring

  Research integrity matters more than tokenization; L2 economics keep cost per analysis well under $0.01.

• RAG over literature

  Biologists need the cited paper, not just a number - RAG keeps the model honest.