Evo2 Variant Intelligence
A research-grade interface for variant pathogenicity prediction, connecting the Evo2 foundation model to a usable analysis console with 3D protein visualization, clinical annotation, and session history.
Overview
Evo2 Variant Intelligence is a full-stack genomics research platform built to make variant pathogenicity prediction accessible and interpretable for researchers. It connects to the Evo2 7B model running on H100 GPU infrastructure via Modal, accepting SNV, deletion, and insertion variants and returning likelihood scores with clinical context.
The platform integrates directly with NCBI for gene search, UCSC for sequence fetching, and ClinVar for variant annotation — giving researchers a single interface for the full analysis workflow. A Molstar 3D viewer renders protein structure context alongside prediction results.
Built with a Next.js frontend, Python Modal backend, and Supabase for session and history management, the platform demonstrates serious backend inference architecture: real GPU compute, external scientific API integrations, and a research-grade UX designed for domain experts.
What we built
Screenshots
The challenge
Connecting a cutting-edge genomic foundation model to a usable research interface required bridging async GPU inference, multiple external scientific APIs, and a complex domain UX — all while keeping the experience fast and interpretable.
The outcome
A production-grade genomics research console with live Evo2 model inference, ClinVar and NCBI integration, Molstar 3D visualization, and a session-persistent analysis history — demonstrating real deep-tech engineering capability.
Technical stack
More work
Want to build something like this?
Have a system, product, campaign, or visual experience that needs building?