Live Product — Agentic AI Inside & Out

Home & Car AI
that thinks before it answers

An AI repair assistant that autonomously checks your repair history, open recalls, and overdue maintenance before responding — and a product built end-to-end using GitHub Copilot agent mode.

Try the App → View on GitHub
1,200+Automated Tests
15DB Tables
20+API Endpoints
111Agent Pitfalls Documented
WeeksIdea to Production

What the App Does

Track every asset you own — vehicles, appliances, home systems — with AI that actually helps.

📸

Photo → Asset Details

Point your camera at any appliance. Claude Vision extracts brand, model, serial number, and type automatically — no forms required.

🤖

Agentic Repair Engine

Ask about a symptom. Claude autonomously decides which tools to call — checking your repair history, open safety recalls, and overdue maintenance — before crafting an answer specific to your asset.

⚠️

Proactive Alerts

Warranty expiration warnings, recall notifications for your vehicles, and maintenance reminders — computed from your actual asset data.

📄

Document Storage

Upload manuals, warranty cards, and receipts. All stored and linked to the right asset — never lose a manual again.

🚘

Vehicle + Home

Cars, trucks, refrigerators, HVAC systems, washers — every asset type in one place with asset-appropriate AI context for each.

📱

Mobile-First

Designed for the garage and the utility room. Bottom navigation, card-based layout, touch-friendly — works great from your phone.

Agentic AI: In the Product & In the Build

Two distinct uses of agentic AI — one running in production answering user queries, one used during development to build the product itself.

🤖 Agentic Repair Engine (IN the Product)

  • Claude uses tool calling to decide what context to fetch before answering
  • get_repair_history — “you had the brakes done 8 months ago”
  • get_open_recalls — “there’s an open recall on your brake booster — it’s covered for free”
  • get_maintenance_status — “your brake fluid is 60 days overdue”
  • 7 total tools: repair history, recalls, maintenance, YouTube, Amazon, stores, contractors
  • Claude decides autonomously which tools to call based on the query

🛠️ Built Using Agentic AI (GitHub Copilot)

  • GitHub Copilot agent mode drove architecture, tests, and deployment
  • Writes tests before implementation (TDD), catches its own errors
  • Handles DB schema → API → frontend → deploy in one session
  • 111 pitfalls documented to prevent repeated mistakes

👁️ Claude Vision API

  • Appliance photo → structured JSON (brand, model, serial)
  • Insurance card → policy number, carrier, vehicles
  • Client-side canvas resize to stay under Lambda's 6MB invocation limit
  • HEIC validation, graceful fallback on unsupported formats

☁️ Serverless on AWS

  • Lambda (Node.js 24.x) + API Gateway HTTP v2
  • Aurora DSQL — IAM-auth PostgreSQL (no SERIAL, no FK constraints)
  • S3 + CloudFront for global frontend delivery
  • Secrets Manager — zero plaintext credentials anywhere
  • Staging gate: smoke tests run before every prod deploy

🛡️ Production Quality

  • 1,200+ Jest unit tests + schema contract validation
  • OWASP Top 10 compliance — rate limiting, brute-force lockout, JWT auth
  • No error.message in API responses (OWASP A05)
  • Email verification + admin approval workflow
  • Automated deploy pipeline with syntax check + test gate
📋

111 Agent Pitfalls Documented

Every production bug, wrong assumption, and wasted debugging hour is documented in the copilot-instructions.md file — 111 numbered pitfalls covering Aurora DSQL quirks, Lambda cold start gotchas, Mailgun credential types, and more. A real-world guide to agentic AI development that ships to production.

Tech Stack

Standard AWS serverless pattern, chosen for operational simplicity and near-zero cost at indie scale.

⚡ AWS Lambda Compute
🌐 API Gateway HTTP v2 API
🐘 Aurora DSQL Database
☁️ S3 + CloudFront Frontend
🔒 Secrets Manager Secrets
🟩 Node.js 24 / Express Runtime
🤖 Claude API (Anthropic) AI
👁️ Claude Vision Multimodal
🧬 Jest + Supertest Testing
✉️ Mailgun Email