CMC Thesis Chatbot

A Retrieval-Augmented Generation (RAG) chatbot that answers questions about Claremont McKenna College senior theses using real database records — not general model knowledge.

🌐 https://cmcthesischatbot.com

---

What It Does

This system translates natural language questions into structured database queries and semantic searches over CMC thesis metadata.

It can:

• Search theses by title, topic, advisor, department, or year
• Rank advisors by topic expertise
• Filter by award, season, or publication date
• Summarize real abstracts
• Generate thesis ideas grounded in actual CMC data

Unlike ChatGPT, every response is backed by records from the CMC thesis archive.

---

Architecture

Backend: Flask
Database: SQLite (theses2.db)
Embeddings: SentenceTransformers (all-MiniLM-L6-v2)
Vector Store: ChromaDB (persistent, cosine similarity)
LLM: Groq llama-3.1-8b-instant
Deployment: Docker + AWS EC2
CI/CD: GitHub Actions

---

Query Pipeline

classify() → fetch() → respond()
  (LLM)        (no LLM)      (LLM)

1. Classify

LLM extracts:

• intent (title lookup, topic search, aggregation, person lookup)
• entities (names, topics, filters)

2. Fetch

Pure retrieval:

• SQL for structured queries
• Vector search (ChromaDB) for semantic topic matching
• Hybrid (SQL filter → vector re-rank) for constrained topic queries

3. Respond

LLM formats only the retrieved records into a grounded answer.
It is explicitly instructed not to use outside knowledge.

---

Hallucination Prevention

• Topic-based advisor rankings use vector search first, then count advisors within real results
• Prompts include strict grounding instructions
• Advisor and author lookups run separate SQL queries

The model never invents advisors or theses.

---

Data

Source: Scholarship@Claremont
Stored in SQLite with fields including:

• Title
• Author(s)
• Advisor(s)
• Department(s)
• Abstract
• Keywords
• Award
• Publication date
• Season
• URL

---

Interface Examples & Comparison With ChatGPT-4o

Thesis Ideas / Outline

CMC Thesis Chatbot	ChatGPT-4o

---

Advisor Search

CMC Thesis Chatbot	ChatGPT-4o

---

Thesis Search

CMC Thesis Chatbot	ChatGPT-4o

---

Project Structure

rag_system17.py      # classify → fetch → respond
config.yaml          # acronym expansion
theses2.db           # SQLite database
chroma_store/        # persistent vector index
screenshots/
Dockerfile
.github/workflows/deploy.yml

---

Setup

pip install chromadb sentence-transformers groq numpy pyyaml flask

# Build vector index (run once)
python rag_system17.py --build-index

# Start chatbot
python rag_system17.py

Set API key:

export GROQ_API_KEY=your_key_here

---

Future Improvements

• Web UI (currently CLI)
• Caching layer
• Full-text thesis embeddings
• Multi-turn conversation support