Technical details

How it works

J Thrust is a local-first FastAPI and React prototype. The core triage logic is rule-based for safety and explainability. AI is used only as a constrained explanation layer.

System flow

CSV upload

The user uploads a CSV from the included demo datasets. The backend accepts the file and reads it with Python/pandas.

Validation

The backend checks that required fields exist, including variant ID, gene, HGVS, current classification, evidence snapshots, and routing signals.

Rule-based evidence routing

The engine checks fields such as splice suspicion, ancestry data gap, affected-relative availability, literature level, and functional evidence relevance.

Priority and actor assignment

The backend assigns priority, primary evidence path, secondary paths, and the suggested lab/research actor.

Frontend report

The React frontend displays the ranked queue, source snapshots, single-variant report, evidence map, and internal lab/research note.

AI-assisted explanation

OpenAI is called only after the rule-based result exists. It explains the already-structured output and is not allowed to classify variants or make recommendations.

Backend stack

FastAPI for backend routes.
pandas for CSV parsing.
uvicorn for local development server.
openai Python SDK for optional explanation generation.

Frontend stack

React for the user interface.
Vite for local development.
Axios for backend API calls.
Custom CSS for the visual system.

Safety boundary

The system does not diagnose, classify variants, determine benign/pathogenic status, or make medical-management recommendations. It only routes uncertain variants toward possible missing-evidence paths.