← Back to projects

Spring AI RAG + Tool-Calling Chatbot

Conversational AI with RAG, agentic tool-calling, and an LLM-as-Judge layer to minimize hallucinations.

Spring AISpring BootReactpgvectorPineconeDockerAWS
View Source

Overview

Master’s capstone at Kennesaw State University — a production-grade conversational AI system that retrieves context via RAG, evaluates response faithfulness with an LLM-as-Judge layer, and autonomously invokes external REST APIs through tool-calling based on user intent. Source code is available on GitHub

Key Features

  • RAG pipeline — Vector embeddings with pgvector and Pinecone for semantic search across large document corpora
  • LLM-as-Judge — Scores and filters responses for faithfulness to reduce hallucinations
  • Tool calling — Agentic function calling turns the chatbot into an action-capable AI agent
  • Full stack — Spring Boot backend, React frontend, Docker, and AWS deployment

Outcome

Addresses a core reliability gap in enterprise AI adoption by measurably reducing hallucinations through automated response evaluation.