Inspiration
To make online shopping feel more natural, helpful, and personalized through conversation.
What it does
It asks smart follow-up questions and recommends the user’s target product from a 50,000-item catalogue within 10 turns.
How we built it
We combined intent detection, rule-based constraint parsing, hybrid BM25/vector retrieval, cross-encoder reranking, and adaptive clarification—running locally without runtime LLM APIs.
Challenges we ran into
Balancing search accuracy, response speed, incomplete preferences, and intent changes during a conversation.
Accomplishments that we're proud of
We improved the TechnicalScore from 0.107 to 0.7836, achieving a 92% Hit Rate@10 and finding products in an average of 3.86 turns.
What we learned
Effective conversational search depends on strong state management, high-quality retrieval, and asking questions that provide useful information.
What's next for TechJam Conversational E-Commerce Search Agent
We plan to make the system fully domain-agnostic, improve free-text understanding,
Log in or sign up for Devpost to join the conversation.