🎓 RAG-Based Course Recommender for Academic Advising

This repository contains the implementation of a Retrieval-Augmented Generation (RAG) system developed to assist university students in selecting the most suitable courses based on reviews, outlines, and historical feedback. The project was built as part of a university course on Speech and Language Processing for Generative AI, using state-of-the-art LLMs and embedding models.

📘 Institution: Lahore University of Management Sciences (LUMS)
🧠 Course: Speech & Language Processing for Generative AI
🗓️ Semester: Spring 2024

---

🧠 Project Overview

The system provides a personalized academic advising chatbot powered by a RAG pipeline. It retrieves relevant course reviews and outlines based on user queries and uses a fine-tuned LLM to provide concise, informed responses about course workload, difficulty, prerequisites, and study suggestions.

The backend integrates:

• Rich historical reviews gathered from the LUMS Discussion Forum and Google Forms
• Manually curated course outlines from the Registrar Portal
• Semantic vector embedding using nomic-embed-text
• Conversational memory using LangChain to support follow-up queries

---

🖼️ Pipeline Overview

---

📁 Directory Structure

RAG-Course-Recommender/
│
├── data/                          # Raw and processed academic data
│   ├── course reviews/           # Collected Google Form + LDF reviews
│   └── vector_store_nomic/       # Precomputed vector embeddings
│
├── flagged/                      # Manually flagged outputs (low quality)
│
├── notebooks/                    # Development notebooks
│   └── main.ipynb
│
├── report/                       # Final submission documents
│   ├── G16-Final_Report.pdf
│   └── G16-Poster.pdf
│
├── Pipeline.jpeg                 # Architectural pipeline visualization
├── requirements.txt              # Python dependencies
├── README.md                     # Project documentation
├── .gitignore
└── .DS_Store                     # (To be deleted)

---

🔧 Technical Stack

Component	Technology Used
Vector Embedding	nomic-embed-text from Ollama
Vector Store	ChromaDB (via LangChain)
Language Model	mistralai/Mistral-7B-Instruct-v0.2
Prompt Template	LangChain PromptTemplate
Memory Chain	LangChain ConversationBufferMemory
Document Loaders	PyPDFLoader, Docx2txtLoader
Interface	Gradio with local + ngrok link sharing

---

✨ Key Features

• 🔍 Semantic Document Retrieval: Matches user prompts with relevant reviews and outlines
• 🧠 LLM-Powered Reasoning: Generates summaries, advice, and answers using context-aware prompts
• 💬 Conversational Memory: Tracks prior queries to support follow-up questions
• 📊 Multi-source Data: Combines formal outlines and student opinions
• 🌐 Gradio Interface: Lightweight, shareable user interface for easy access

---

🚀 Getting Started

Prerequisites

• Python 3.9+
• pip or conda

Installation

git clone https://github.com/SaadH-077/RAG-Course-Recommender.git
cd RAG-Course-Recommender
pip install -r requirements.txt

---

▶️ Running the Application

You can launch the interactive chatbot through the Gradio interface by running:

python main.py

This will start a local server (and optionally a public link via ngrok).

---

📄 Final Deliverables

• 📘 G16-Final Report (PDF)
• 📊 G16 Poster Presentation (PDF)

These documents contain detailed methodology, model comparisons, results, and future work.

---

📌 Citation

If you reference this project in academic or practical work:

@project{RAGCourseAdviser2025,
  title = {RAG-Based Course Recommender for Academic Advising},
  note = {Developed for Speech & Language Processing for Generative AI, Spring 2025, LUMS}
}

---

“Retrieval grounds generation — context is not optional, it's everything.”