Qinxuan Li
← All Work

AI product

Unifeast

A multi-turn dining recommendation chatbot that uses retrieval and rule-based allergen filtering outside the language model.

Role
AI/ML Engineer, Amazon University Engagement Programme
Status
Prototype
Year
2025

Overview

A conversational dining recommendation prototype built during the Amazon University Engagement Programme.

Background

The prototype explored multi-turn dining recommendations while treating allergen safety as a system constraint rather than a generated suggestion.

Response

As sole developer, Qinxuan built three backend iterations, moving from the Bedrock console to LangChain. The final design keeps a rule-based allergen filter outside the language model.

Methods

Conversational AI · Retrieval-augmented generation · Rule-based allergen filtering · Prototype evaluation

Outcome

The project has a four-tab mobile demo, sample data, a small RAG implementation and an evaluation record.

Recognition

  • Best Technical Implementation

Project materials

Project document being prepared for release.