Axle

Formerly CarScout — renamed August 2026.

Live

• App: https://d1j3m9qdbgs5ik.cloudfront.net

Architecture

CloudFront + S3 (React/Vite SPA)
        |
        v
API Gateway (HTTP API) -> Lambda (FastAPI via Mangum, python3.12, arm64)
        |
        +-- DynamoDB  carscout-listings-cache   shared listing cache, native TTL
        +-- Supabase  auth + saved vehicles
        +-- Auto.dev  live inventory (1,000 calls/month free tier)

The listing cache is three tiers: in-process memory (900s) -> DynamoDB (shared, 30d) -> local disk -> Auto.dev. On Lambda the disk tier lives in /tmp, which is per-instance and wiped on cold start; DynamoDB is what actually protects the monthly API quota across invocations.

Deploying

# API
pip install -r backend/requirements.txt --platform manylinux2014_aarch64         --implementation cp --python-version 3.12 --only-binary=:all: --target build_pkg
cp backend/*.py build_pkg/ && cp -r data build_pkg/data
sam deploy --stack-name carscout --resolve-s3 --capabilities CAPABILITY_IAM

# Frontend
cd frontend && npm run build
aws s3 sync dist/ s3://carscout-web-<account-id>/ --delete

One-time table setup: python backend/create_cache_table.py

AI-powered used-car shopping. Describe what you want in plain English, and CarScout finds, ranks, and compares real listings from nearby dealers.

Originally a TAMUhack proof-of-concept (hardcoded to a single Toyota dealership), now being rebuilt into a real, deployed, nationwide product. See docs/ROADMAP.md for the architecture and plan.

How it works

A chat-style UI collects your zip code and preferences. The backend uses an LLM to turn natural language into structured preferences, then scores and ranks inventory and generates AI-powered comparisons between vehicles.

Tech stack

• Frontend: React 18 + Vite (frontend/)
• Backend: FastAPI + OpenRouter LLM (backend/)
• Data (current): static data/carData.json
• Data (target): Supabase Postgres, fed by an ingestion pipeline (Auto.dev listings + NHTSA VIN decode + LLM enrichment) — see the roadmap.

Local development

Prerequisites: Node ≥ 18, Python ≥ 3.11.

Backend

cd backend
python -m venv .venv && . .venv/Scripts/activate   # Windows; use .venv/bin/activate on macOS/Linux
pip install -r requirements.txt
cp .env.example .env        # then add your OPENROUTER_API_KEY
python -m uvicorn main:app --reload --port 8000

Frontend

cd frontend
npm install
cp .env.example .env        # VITE_API_BASE defaults to http://localhost:8000
npm run dev

Then open the Vite dev URL (default http://localhost:5173).

npm start (from frontend/) runs the frontend and backend together via concurrently.

Project structure

backend/     FastAPI app (preference extraction, scoring/ranking, compare)
frontend/    React + Vite chat UI
data/        carData.json + scrapers (being replaced by the ingestion pipeline)
docs/        ROADMAP.md and design notes

Status

Active rebuild. The current main still runs on the static demo dataset while the Supabase-backed ingestion pipeline is built out per the roadmap.