Human Debug - AI for genomics

Problem in a nutshell
Add new patient
Upload medical history
Extract medical history
Extract structured medical data
Automated, interactive DNA analysis
Draft genomic diagnostic report

Inspiration

Previously built preventive radiology AI startup Floy. Realized that genetic diagnostics is bottlenecked by slow, costly, and error-prone analytics.

3 months waiting time for result
$1,200 costs in analysis alone (excluding sequencing)
Critical shortage of 2,640 genetic counsellors
Long NHS waiting list from family

A key challenge is the integration of patient history, genetic analysis, and latest scientific research.

What it does

AI platform to help medical doctors perform genetic diagnostics 5x faster and potentially 50% cheaper. Doctors create a new patient case and upload the scanned/photographed medical history letter from which structured data (phenotypes) is extracted using the Pixtral model. Then, they upload the sequenced patient genome which automatically processed and any variants highlighted for the doctor's evaluation. For highly specific questions, a chat interface with PubMed search provides easy access to the latest biomedical research papers. The completed analysis is then automatically turned into draft diagnostic report.

How we built it

Frontend: React/NextJS
Backend: Python, fastAPI, supabase, freebayes, bwa, samtools, bcftools
Data: Reference Genome GRCh38 - hg38

Challenges we ran into

Working first time with genomics data. Learning everything about how to get and process it
Learning about genomics and the diagnostic problems

Accomplishments that we're proud of

Building a fully functional & deployed prototype! -> https://human-debug.vercel.app/
Creating a well-design and interactive UI that addresses key workflow needs of doctors