2025 NASA Space Apps Challenge

A World Away: Hunting for Exoplanets with AI

Group: vibe coding only

• Nick
• Carl
• Mahad
• Lynn
• Tahmid
• Javier

Getting started

This repository contains a small Flask backend (model inference) and a React + TypeScript frontend. Below are concise, copy-paste PowerShell instructions to get both running on Windows. These steps assume you have Python 3.10+ and Node.js installed.

Backend (Flask + TensorFlow)

1. Open PowerShell and change to the backend folder:

cd .\backend

2. Create a virtual environment and activate it (recommended):

python -m venv .venv
. .venv\Scripts\Activate.ps1

3. Install Python dependencies. If a requirements file isn't present, install the essentials used by the project:

pip install --upgrade pip
pip install flask flask-cors flask-socketio tensorflow numpy

4. Make sure the model file exoplanet_lightcurve_cnn.h5 exists in the backend/ folder. The /predict endpoint expects JSON with fields time (CSV string), flux (CSV string), period, and t0.

5. Run the backend server (development):

# from backend folder
py app.py

Frontend (React + TypeScript)

1. Open a new PowerShell window and change to the frontend folder:

cd ..\frontend

2. Install node dependencies:

npm install

3. Start the dev server:

npm start

By default the dev server runs on http://localhost:3000

Testing / testcases

• Useful test inputs are in backend/testcases/ (JSON files). Each file contains time, flux (both CSV strings) and period, t0 fields. You can POST these files directly to the backend for a quick check.

Example PowerShell POST (send a saved testcase to the backend):

$body = Get-Content .\backend\testcases\kepler227_synth.json -Raw
Invoke-RestMethod -Uri http://localhost:5000/predict -Method Post -Body $body -ContentType 'application/json'

If you get a CORS or network error when the frontend tries to call the backend, verify both servers are running and check the browser console (DevTools) for the actual error.

Notes and troubleshooting

• The backend's preprocessing mirrors the training code: flux is median-centered and scaled by std, the time domain is phase-folded using period and t0, then interpolated to the model input size (501 samples). If you modify preprocessing in backend/app.py, re-check parity with backend/ai_model_training/data_processing.py (training code).
• If the model returns unexpectedly confident outputs on tiny inputs, provide longer, higher-sample testcases (the backend/testcases/ folder contains some exported Kepler examples and a synthesized confirmed-like transit kepler227_synth.json).
• To enable detailed debugging in the backend (raw logits, preprocessed array), update /predict to return additional fields guarded by a query parameter (e.g. ?debug=1).

If anything is unclear or you'd like me to add an automated script to start both servers (or to create a requirements.txt / package.json scripts), tell me and I can add it.