Building the Solution
We engineered an AI-powered mechanic assistant using Claude's tool-use capabilities, built on a Next.js full-stack architecture and deployed on Vercel with Supabase as the data layer.
Intelligent Vehicle Identification
We built a multi-country license plate lookup system that connects to manufacturer databases across 9 countries. Users enter their plate number, and the system automatically resolves make, model, year, and internal vehicle identifiers — eliminating manual data entry and unlocking access to vehicle-specific repair data from the first interaction.
AI Agent with 24 Specialized Diagnostic Tools
At the core of the product is a Claude-powered agent (Sonnet 4.6) equipped with 24 automotive tools. For example:
- engine oil specs
- brake data
- camshaft drive replacement procedures
- diesel aftertreatment diagnostics
Claude autonomously decides which tools to invoke based on the user's question, chaining up to 10 tool calls per turn for complex multi-step diagnoses. When vehicle-specific data isn't available, the agent falls back to web search, ensuring no question goes unanswered.
Transparent Streaming diagnostics
We implemented a real-time Server-Sent Events streaming pipeline that displays exactly howuser’s diagnosis was reached. This transparency builds trust with DIY owners making safety-critical repair decisions.
Cost-efficient Prompt Caching
To keep the product economically viable at scale, we enabled Anthropic's prompt caching on system prompts and conversation history. This significantly reduces token costs on follow-up messages within a conversation, where the vehicle context and prior exchanges are already cached.
Token-based Usage Metering
Rather than a blunt message count, we built a granular token-based billing system that logs usage per tool-calling leg. Simple questions cost less than complex multi-tool diagnostics, giving users fair value across a Free, Plus, and Pro tier — with Stripe handling subscriptions and billing management.
