ECE Classroom Assistant — RAG-Based Domain-Specific QA System

A Dynamic Domain-Specific Classroom Assistant built using Retrieval-Augmented Generation (RAG) architecture. Designed for Electronics and Communication Engineering (ECE) departments to query student records and academic documents using natural language.

---

Project Overview

Traditional LLMs generate responses from generalized training data and often hallucinate. This system solves that by grounding every answer strictly in uploaded classroom documents — student records, syllabi, and subject notes.

Key capabilities:

• Answer student-specific queries from uploaded CSV records
• Retrieve subject knowledge from uploaded ECE PDF documents
• Reject out-of-domain queries with a controlled response
• Dynamically update knowledge base without retraining the model

---

System Architecture

User Query
    ↓
Embedding Model (all-MiniLM-L6-v2)
    ↓
Vector Similarity Search (ChromaDB)
    ↓
Top-K Relevant Chunks Retrieved
    ↓
Prompt Construction (query + context)
    ↓
LLM Response Generation (phi3:mini via Ollama)
    ↓
Grounded Answer → User

---

Tech Stack

---

Project Structure

rag_classroom_assistant/
├── app.py                  # Streamlit chat UI
├── requirements.txt        # All dependencies
├── .env                    # Environment config
├── verify_setup.py         # Setup checker
│
├── src/
│   ├── ingest.py           # Document ingestion + chunking
│   ├── embed.py            # Embedding generation + ChromaDB storage
│   └── rag_pipeline.py     # Query → retrieve → LLM → answer
│
├── data/
│   ├── documents/          # ECE PDFs (syllabus, notes)
│   └── students/           # Student records (CSV)
│
└── db/
    ├── chunks.json         # Processed text chunks
    └── chroma_store/       # ChromaDB vector index

---

Setup Instructions

1. Clone / download the project

cd Desktop
# place the rag_classroom_assistant folder here

2. Create and activate virtual environment

python -m venv venv
venv\Scripts\activate        # Windows

3. Install dependencies

pip install -r requirements.txt

4. Install Ollama and pull model

Download Ollama from https://ollama.com then:

ollama pull phi3:mini

5. Add your documents

• Place ECE PDFs in data/documents/
• Place student CSV in data/students/

6. Run ingestion and embedding

python src\ingest.py
python src\embed.py

7. Launch the app

streamlit run app.py

---

Usage

Open http://localhost:8501 in your browser and start asking questions:

Student queries:

• "What are Priya's marks in VLSI Design?"
• "Who scored the highest in Digital Signal Processing?"
• "List students with attendance below 80%"

Subject queries (requires PDFs uploaded):

• "What is the syllabus for Microprocessors?"
• "Explain the topics covered in VLSI Design unit 2"

Out-of-domain (correctly rejected):

• "What is the capital of France?" → "This information is not available in the uploaded documents."

---

How RAG Works (for interviews)

1. Ingestion — Documents are read and split into 500-word overlapping chunks
2. Embedding — Each chunk is converted to a 384-dimensional vector using sentence-transformers
3. Storage — Vectors are stored in ChromaDB with cosine similarity indexing
4. Retrieval — User query is embedded and top-3 most similar chunks are fetched
5. Generation — Retrieved chunks + query are sent to phi3:mini as a structured prompt
6. Domain restriction — If similarity score < 0.30, query is rejected as out-of-domain

---

Why RAG over plain LLM?

---

Future Enhancements

• Add multi-language support (Tamil, Hindi)
• Integrate with college ERP system for live student data
• Add voice input support
• Deploy on cloud (AWS / Azure) for department-wide access
• Add answer confidence scores visible to the user

---

Author

Vashundhara.R — Pre-final year ECE Student
Project: AIML mini Project
Domain: AI / NLP / Information Retrieval