University-Rag-Assistant

🎓 University RAG Assistant

An AI-powered document intelligence system for Stevens Institute of Technology
Ask questions about admissions, tuition, deadlines, courses, and international student policies —
and get cited, conflict-aware answers grounded strictly in official university documents.

---

✨ What It Does

This project turns five official Stevens Institute PDF documents into a queryable knowledge base. You ask a question in natural language; the system retrieves the most relevant chunks, sends them to an LLM, and returns a sourced, structured answer.

The standout feature is conflict detection — if two documents disagree on the same fact (e.g. two different application deadlines), the app doesn't silently pick one. It:

1. 🟢 Highlights the preferred answer from the more recently published source
2. 🔴 Flags the conflict with both values and their sources side-by-side
3. Recommends the student verify directly with the relevant office

---

📂 Project Structure

University-Rag-Assistant/
│
├── app.py                    # Streamlit UI — two-column layout, conflict cards
│
├── src/
│   ├── __init__.py
│   ├── rag.py                # Retrieval + LLM answering + conflict-aware prompt
│   └── ingest.py             # PDF loading, sentence-aware chunking, ChromaDB upsert
│
├── data/                     # Drop your PDF documents here
│   ├── 01_Stevens_Admissions_Guide.pdf
│   ├── 02_Stevens_Course_Catalogue.pdf
│   ├── 03_Stevens_Tuition_Fees.pdf
│   ├── 04_Stevens_Academic_Calendar.pdf
│   └── 05_Stevens_International_FAQ.pdf
│
├── chroma_db/                # Auto-created by ingest.py (git-ignored)
├── .env                      # Your API keys (git-ignored)
├── .gitignore
├── requirements.txt
└── README.md

---

🧠 Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         INGEST PIPELINE                         │
│                                                                 │
│  PDF files  →  PyPDF page extract  →  Sentence-aware chunker   │
│            →  all-MiniLM-L6-v2 embeddings  →  ChromaDB upsert  │
└─────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────┐
│                          QUERY PIPELINE                         │
│                                                                 │
│  User question  →  Embed query  →  ChromaDB top-8 retrieval    │
│               →  Build context (with chunk + page labels)       │
│               →  Conflict-aware prompt  →  Groq LLaMA 3.3 70B  │
│               →  Parse ✓ / ⚠ markers  →  Render styled cards   │
└─────────────────────────────────────────────────────────────────┘

Component	Technology	Why
Vector store	ChromaDB (persistent)	Local, fast, no infra needed
Embeddings	all-MiniLM-L6-v2	384-d semantic vectors, free
LLM	LLaMA 3.3 70B via Groq	Strong multi-source reasoning, free tier
Chunking	Sentence-boundary aware	Dates & deadlines never split mid-sentence
UI	Streamlit + custom CSS	Two-column, dark academic theme

---

🚀 Getting Started

1 — Clone the repo

git clone https://github.com/yourusername/University-Rag-Assistant.git
cd University-Rag-Assistant

2 — Create and activate a virtual environment

python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
source .venv/bin/activate

3 — Install dependencies

pip install -r requirements.txt

4 — Add your API key

Create a .env file in the project root:

GROQ_API_KEY=your_groq_api_key_here

Get a free key at console.groq.com.

5 — Add your PDFs

Drop your university PDF documents into the data/ folder.

6 — Ingest documents

python src/ingest.py

This extracts text, chunks it sentence-by-sentence, generates semantic embeddings, and stores everything in ChromaDB. Re-run any time you add or update documents — upsert makes it safe to re-run.

7 — Launch the app

streamlit run app.py

Open http://localhost:8501 in your browser.

---

💡 Example Questions to Try

Question	Expected behaviour
What is the priority deadline for graduate master's programs?	🔴 Conflict detected — two different dates found across document sections
What is the minimum TOEFL score for international students?	🟢 Clean answer — 79 iBT, cited from International FAQ
How much is undergraduate tuition for 2025–2026?	🟢 Clean answer — $62,428, cited from Tuition & Fees doc
When does the fall 2025 semester start?	🟢 Clean answer — August 28, 2025, from Academic Calendar
Can F-1 students work off-campus in their first year?	🟢 Clean answer with policy details from International FAQ

---

⚠️ Conflict Detection in Action

When two sources disagree, the UI renders two distinct cards:

┌─────────────────────────────────────────────────────┐
│ ✓  PREFERRED ANSWER (most recent source)            │
│                                                     │
│  Based on Source 1 (Admissions Guide, Aug 2024):    │
│  The priority deadline is February 1, 2025.         │
└─────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────┐
│ ⚠  CONFLICT DETECTED                               │
│                                                     │
│  Source 3 (Admissions Guide, page 2, chunk 2)       │
│  states: January 15, 2025.                          │
│                                                     │
│  Please confirm with the Office of Admissions.      │
└─────────────────────────────────────────────────────┘

The LLM follows a strict 5-rule resolution protocol — it never silently picks a value when sources disagree.

---

🛠 Troubleshooting

OpenBLAS memory error on Windows

$env:OPENBLAS_NUM_THREADS = "1"
$env:OMP_NUM_THREADS = "1"
python src/ingest.py

src module not found Make sure src/__init__.py exists and you're running commands from the project root, not inside src/.

Conflict not detected Ensure build_context() in rag.py is NOT deduplicating chunks by page — two chunks from the same page can still conflict and must appear as separate sources.

Re-ingesting after doc changes Just re-run python src/ingest.py — upsert() handles duplicates safely.

---

📦 Requirements

streamlit
chromadb
pypdf
sentence-transformers
groq
python-dotenv

Install all at once:

pip install streamlit chromadb pypdf sentence-transformers groq python-dotenv

---

🗺 Roadmap

• Multi-university support (swap document sets via sidebar)
• Source confidence scores displayed per chunk
• PDF viewer panel — click a source chip to see the original page
• Chat history — multi-turn conversation with memory
• Export answers as PDF report
• Admin panel — drag-and-drop document upload + re-ingest trigger

---

🤝 Contributing

Pull requests are welcome. For major changes please open an issue first to discuss what you'd like to change.

1. Fork the repo
2. Create your branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

---

📄 License

Distributed under the MIT License. See LICENSE for more information.

---

⬆ Back to top