2nd Place at AGI House Parametric Memory Build Day

We built Mutation Memory: ChatGPT for DNA sequence data, and took 2nd place at AGI House's Parametric Memory Build Day hackathon.

The idea

Geneticists work with massive amounts of DNA sequence data, but querying it has always been clunky (think custom scripts, rigid search tools, and a lot of context-switching). We wanted to make it as easy as asking a question in plain English.

Mutation Memory is a chat interface that lets researchers query DNA data conversationally. Ask it about a sequence, a mutation, a gene region and it retrieves relevant results from a vector database of DNA sequence and text embedding pairs.

How it works

Under the hood, we use the Nucleotide Transformer to generate embeddings from raw DNA sequences, then store those alongside text embeddings in MongoDB. When a user asks a question, the system finds the most relevant sequences and surfaces them through a chat interface powered by ChatGPT and Ollama.

The whole thing came together in a single day using Windsurf as our development environment.