SAGA

SAGA turns public haemophilia updates into short, source-linked briefs for human review. Each brief explains what changed, why it may matter, who should review it, and a possible next step.

Built by Team Codechondria, Amity University, Bengaluru, for the Novo Nordisk GBS Hackathon 2026.

Start here

• Open the live demo. No sign-in or installation
• Start the guided walkthrough
• Presentation · PDF (not included in this installation)
• Watch the demo (2:55) · Demo guide
• Technical review and app download. Source and classifier included

Run locally

Install Docker Compose, uv and OpenSSL. Extract the app download, then run:

cd SAGA/source
./scripts/setup_developer.sh

Open localhost:3000. Initial setup needs internet; replay needs no provider key. For a Git clone, the same script downloads the model through an authenticated GitHub CLI.

How it works

A local classifier sorts records. A language model drafts briefs. Evidence checks, routing rules and human review control what appears. The demo uses public records and clearly marked synthetic examples. Its results show the workflow; human usefulness and time savings still need a pilot.

Category accuracy was 92.41% versus the keyword baseline's 55.70% on 79 AI-labelled records. Source-change detection tied its rules baseline. Safety recall was 2/5; see Results.

Guides

Architecture · Results · Models · Data · Sources · Pilot · Operations · Development

Code is in src/saga/ and web/; tests are in tests/. The repository is private. GitHub links require access.

For product context, see existing approaches and SAGA’s focus. The comparison distinguishes published vendor capabilities from our own measured baseline results.