Decide who this project is for, what problem it solves, and list the exact features needed for version 1.

• •Target User & Problem: Clearly define who needs this and the pain point it removes
• •MVP Scope: Accept historical flight data & weather feeds -> Process -> Return verified clean output
• •Ship Goal: Working React UI + FastAPI backend + Supabase DB deployed live online

Create the code repository with React on the frontend and FastAPI on the backend.

• •Project structure: /frontend (React UI) and /backend (FastAPI Python API)
• •Install starter packages: httpx, pydantic, supabase, dotenv, pytest, lucide-react
• •Create .env file for secret keys, add setup instructions to README, and make first Git commit

Build the visual interface in React with clean input forms, loading spinners, and result screens.

• •Sliders and inputs to test different feature values and numbers
• •Color-coded risk meter (Low, Medium, High risk probability)
• •Simple bar chart showing which factors influenced the prediction most
• •AI action recommendation card with one-click export

Build the FastAPI backend routes and test them with sample data connected to the frontend.

• •POST /api/v1/predict — Run the ML model on input numbers and return risk score
• •POST /api/v1/batch-predict — Upload CSV to get risk scores for multiple rows
• •GET /api/v1/model/metrics — View model accuracy scores and top factor weights
• •POST /api/v1/recommendation/generate — Ask AI for advice based on risk level

Create database tables in Supabase so users can safely save, load, and manage their data.

• •prediction_logs — Stores input values, predicted percentage, and risk level
• •model_versions — Stores model version name, accuracy score, and file path
• •action_playbooks — Stores AI-generated advice and action checklists

Make the app work end-to-end without AI first (save to database, load records, and display on screen).

• •Complete save-and-load flow working smoothly before adding AI complexity
• •Helpful error messages if inputs are missing, invalid, or connections drop
• •Real database records displayed on screen with zero fake placeholder text

Write and test AI prompts to ensure the AI returns clear, reliable answers in clean structured format.

• •Clear system instructions with role definition and good example demonstrations
• •Force the AI to output clean structured data instead of messy free-form text
• •Optimize prompt length to keep response times fast and reduce token costs

Train XGBoost delay classifier, serve via FastAPI, and chain prediction output to LLM travel advisor.

• •Core AI Flow: Run trained ML model on input numbers -> Calculate risk score -> AI writes tailored recommendations based on the score.
• •Data checker that verifies AI responses before showing them to the user
• •Live 'Thinking / Generating...' status indicator on screen during processing

Make the AI dependable by adding automatic retries and handling slow or failed requests gracefully.

• •Automatic retry if the AI service is slow or briefly reaches a rate limit
• •Safe backup fallback if the AI returns an unexpected or incomplete answer
• •Tuned prompt instructions for tricky edge cases and long inputs

Run 10–20 realistic test cases to measure how accurate and dependable the app is. Benchmark target: ML ROC-AUC / F1-Score & LLM advisory safety rate.

• •Test Suite: Historical dataset testing: confirm model accuracy score and verify predictions make sense on edge cases.
• •Automated test runner that scores accuracy on every test sample
• •Review failed tests and tune prompts/logic until all key tests pass

Protect the app against spam, malicious inputs, trick prompts, and secret API key leaks.

• •Validate all input numbers to prevent invalid or out-of-range data crashes
• •Block malicious dataset file uploads
• •Secure model file loading so only trusted model weights can run

Clean up the user interface, speed up slow parts, and write clear setup instructions in the README.

• •FastAPI middleware to log request speed, token costs, and error rates
• •Simple README documentation with architecture diagram and step-by-step setup guide
• •UX Polish: smooth copy buttons, keyboard shortcuts, and clean mobile phone layout

Launch the frontend on Vercel and the backend on Railway with a live public web link.

• •Frontend: Live production build deployed to Vercel
• •Backend: Live containerized API deployed to Railway with health check (/healthz)
• •Database: Live Supabase production database connected, secured, and verified

SHIP DAY 🚀: Publish your project, share the link with others, and record a 5-minute demo video.

• •Public Launch: Live website link + Railway API + open-source GitHub repository
• •5-Minute Video: Problem -> Live Demo -> How it works -> Test Results -> What's next
• •Share your launch post and demo video on LinkedIn and Twitter/X