FundFirst backend for Lovable

This package serves the real FundFirst Logistic Regression pipeline through a small FastAPI service. The deployed pipeline reproduces the final FundFirst training procedure: StandardScaler followed by balanced LogisticRegression, fitted on the 2014–2023 development data.

Package contents

FundFirst_backend/
├── main.py
├── requirements.txt
├── render.yaml
├── README.md
└── models/
    ├── fundfirst_logistic_regression.joblib
    └── fundfirst_model_metadata.json

main.py loads the fitted pipeline once at startup and provides /health, /metadata, and /predict. The models folder contains the fitted scaler and classifier together in one joblib pipeline, plus the metadata written according to the notebook.

Run locally

Python 3.12 is recommended.

cd FundFirst_backend
python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
uvicorn main:app --reload

Open http://127.0.0.1:8000/docs, expand POST /predict, choose Try it out, and submit:

{
  "AveragePrice": 515000,
  "MedianAnnualPay": 35500,
  "SavingRatio": 9.0,
  "BaseRate": 4.25
}

Or test from a terminal:

curl -X POST http://127.0.0.1:8000/predict \
  -H "Content-Type: application/json" \
  -d '{"AveragePrice":515000,"MedianAnnualPay":35500,"SavingRatio":9.0,"BaseRate":4.25}'

The response contains the model's class and class probabilities. Raw values must be sent exactly as shown; do not scale them in Lovable because scaling is already inside the saved pipeline.

Deploy on Render

1. Extract this ZIP and upload the FundFirst_backend folder to a new GitHub repository.
2. In Render, select New > Blueprint and connect that repository. Render reads render.yaml automatically.
3. Deploy the service and wait for /health to report {"status":"ok", ...}.
4. In Render, replace the ALLOWED_ORIGINS value * with the exact deployed Lovable origin, for example https://your-project.lovable.app. Multiple origins can be comma-separated.
5. Copy the Render service URL, for example https://fundfirst-api.onrender.com.

For a manual Render web service, use:

• Runtime: Python 3
• Build command: pip install -r requirements.txt
• Start command: uvicorn main:app --host 0.0.0.0 --port $PORT
• Health check path: /health

Connect Lovable

Set Lovable's VITE_API_URL environment variable to the deployed backend URL without a trailing slash. On form submission, send a POST request to ${VITE_API_URL}/predict with:

{
  "AveragePrice": 515000,
  "MedianAnnualPay": 35500,
  "SavingRatio": 9.0,
  "BaseRate": 4.25
}

Use prediction and probabilities from the response as the only prediction source. Do not recreate the model, its scaling, or the TSM thresholds in the frontend.

Important

The pipeline reproduces the project's rule-generated deposit-feasibility labels. It does not determine mortgage eligibility, creditworthiness, or mortgage approval and does not provide regulated financial advice.