FitSpec

Overview

A full-stack automotive parts fitment and recommendation platform that helps customers find verified-compatible parts for their specific vehicle. Features cascading Year/Make/Model/Trim selection across 1,747 vehicles, ML-powered cross-category recommendations, an AI fitment assistant, real-time inventory via SignalR, and a polyglot persistence layer with SQL Server for relational fitment data and MongoDB for flexible product metadata.

Key Features

• Cascading vehicle selector (Year/Make/Model/Trim) with URL-persisted state across 1,747 vehicles and 174K+ fitment mappings
• ML.NET matrix factorization recommendation engine trained on 4,500 orders for cross-category product suggestions
• Claude-powered AI Fitment Assistant with dynamic context injection for vehicle-specific parts Q&A
• Polyglot persistence: SQL Server with Dapper stored procedures for performance-critical fitment queries, MongoDB for flexible product metadata and reviews
• Real-time inventory updates via SignalR push notifications without polling
• Full product lifecycle: comparison tray, install guides, cost estimator, VIN decoder, spec sheets, and SHA256-verified fitment certificates

Challenges & Solutions

The project addressed complex architectural challenges including designing a polyglot persistence strategy with SQL Server for deeply relational fitment data (stored procedures via Dapper for sub-100ms lookups across 174K+ mappings) and MongoDB for semi-structured reviews and variable product specs, building an ML.NET matrix factorization recommendation engine that captures cross-category purchase patterns from simulated order data, orchestrating real-time inventory updates through SignalR hubs, managing AI context for the Claude-powered fitment assistant with dynamic system prompts incorporating vehicle details and compatible products, and containerizing the full stack with Docker Compose health checks to ensure database readiness before API startup.