TH26-Toyota

Inspiration

Many car marketplaces fail when searches are too specific, showing “No results found” even when reasonable alternatives exist. Inspired by how real car salespeople suggest substitutes and explain trade-offs, we set out to build a system where every search returns a meaningful result.

What We Built

An inventory-aware, AI-powered recommendation engine that understands user intent, grounds responses in real inventory, and intelligently relaxes constraints to avoid empty results.

How It Works

Inventory data is loaded from a live dataset
User intent is interpreted using semantic reasoning
Retrieval-Augmented Generation (RAG) matches queries against real inventory
When no exact match exists, constraints are relaxed and substitutes are recommended
Each result includes a clear explanation for transparency

Key Features

Semantic Search: Matches intent, not just keywords
Zero-Match Resilience: Never returns empty results
Comparable Models: Suggests similar alternatives
Budget Awareness: Treats price as a soft constraint
Explainability: Human-readable reasoning for every recommendation

What We Learned

We learned that eliminating empty searches significantly improves user trust and engagement, and that simpler, inventory-grounded architectures can outperform complex systems.

Challenges

Balancing flexible recommendations with strict inventory grounding, optimizing context size, and designing meaningful fallback logic were our main challenges.