University Course Registration AI Agent

An intelligent, terminal-based conversational agent built in Go that helps students navigate complex university course schedules using an Agentic Workflow and Semantic Search.

Project Overview

This project is a CLI-based AI assistant designed to streamline the course discovery process. Moving beyond traditional, static RAG (Retrieval-Augmented Generation) pipelines, this system implements a modern Agentic Architecture. It utilizes OpenAI's Tool Calling capabilities to dynamically orchestrate semantic database queries, handle fuzzy data matching, and even execute local system commands (e.g., drafting emails) based on natural language multi-turn conversations.

Core Engineering Highlights

• Agentic Workflow via Tool Calling: Replaced hard-coded prompts with strict JSON Schema tool definitions (extract_name, query_courses, open_email). The LLM acts as a reasoning engine, deciding when and how to invoke local Go functions.
• Stateful Multi-Turn Dialogue: Implemented a robust context-management system within the CLI loop, allowing the agent to remember previous turns and handle complex follow-up queries seamlessly.
• Semantic Search & Fuzzy Matching: Integrated ChromaDB for vector storage. To handle real-world "dirty data", the system implements a custom fuzzy-matching scoring algorithm to bridge the gap between user typos and canonical instructor names in the database.
• "LLM-as-a-Judge" Unit Testing: Built an automated testing suite (vectorQuery_test.go) that uses an LLM evaluator to assert the semantic accuracy of the Agent's responses against expected outputs, ensuring production-level reliability.

Tech Stack

• Language: Go (Golang 1.23)
• AI/LLM: OpenAI API (go-openai), GPT-4o-mini, Function Calling
• Vector Database: ChromaDB (Dockerized)
• Data Processing: Custom CSV parser & embedding generator

---

Getting Started (Local Development)

1. Prerequisites

• Go (1.23 or later)
• Docker Desktop
• An OpenAI API Key

2. Environment Setup

Create an api.env file in the root directory and add your OpenAI API key:

OPENAI_PROJECT_KEY=your_openai_api_key_here

3. Start the Vector Database

Launch the persistent ChromaDB instance via Docker Compose:

docker-compose up -d

4. Data Ingestion & Execution

To parse the courses.csv, generate embeddings, and start the interactive chat interface, run:

# Use the -reset flag for the first run to build the ChromaDB collections
go run . -reset

# For subsequent runs (faster, skips ingestion):
go run .

---

Testing

The project includes an automated test suite verifying the RAG retrieval accuracy and the Agent's tool-calling logic. Run the tests using:

go test -v

Repository Structure

• main.go - Entry point and stateful Agent dialogue loop.
• db.go - ChromaDB client initialization and collection management.
• query.go - Tool definitions (JSON Schema) and OpenAI tool execution logic.
• insertCollections.go - Batch processing and embedding logic for CSV ingestion.
• vectorQuery_test.go - LLM-as-a-Judge unit tests.