SuperiorInsight

Conversational AI System for Superior University

SuperiorInsight is a production-grade Retrieval-Augmented Generation (RAG) system purpose-built for Superior University. It enables students, faculty, and administrative staff to query complex institutional information — degree programs, fee structures, admission policies, scholarships, and campus facilities — through a natural language interface with sub-second response times.

---

Table of Contents

• Overview
• Technical Architecture
• Key Features
• Project Structure
• Installation
• Configuration
• Running the Application
• Deployment
• Dataset
• Limitations
• Author

---

Overview

Traditional university information portals require users to navigate fragmented documentation across multiple pages. SuperiorInsight consolidates over 4,750 pages of Superior University's digital ecosystem into a single intelligent retrieval layer, enabling accurate, contextual, and multilingual responses without hallucination.

The system is designed with strict domain grounding — it does not speculate beyond the provided knowledge base — and maintains conversational context across multi-turn interactions without topic bleeding.

---

Technical Architecture

User Query
    |
    v
Query Translation Layer (Roman Urdu / Hindi -> English)
    |
    v
History-Aware Retriever
    |
    v
FAISS Vector Database  <---  HuggingFace Embeddings
    |
    v
Context Assembly
    |
    v
Llama-3.3-70B via Groq LPU
    |
    v
Structured Response

Component	Technology
Backend Framework	Flask
Embedding Model	HuggingFace Sentence Transformers
Vector Store	FAISS (Facebook AI Similarity Search)
Language Model	Llama-3.3-70B
Inference Engine	Groq LPU
Orchestration	LangChain
Frontend	HTML / CSS / JavaScript
Deployment Target	Render / Railway / Heroku

---

Key Features

Retrieval-Augmented Generation All responses are grounded in retrieved source documents. The system does not generate information outside the indexed knowledge base.

Multilingual Query Support Handles queries in English, Roman Urdu, and Roman Hindi. A translation layer normalizes non-English input before retrieval, ensuring consistent embedding-space alignment.

History-Aware Retrieval A custom memory engine maintains conversational context across turns. The retriever reformulates follow-up questions using prior context, eliminating the need for users to repeat themselves.

Context Bleeding Prevention Each conversational session operates within strict topic boundaries. Context from unrelated prior exchanges does not contaminate current retrieval.

Groq LPU Inference Responses are generated using Groq's Language Processing Unit, delivering low-latency completions suitable for production environments.

Domain-Locked Prompting A strict system prompt constrains the model to answer only from retrieved context, with a calibrated fallback for out-of-scope queries.

---

Project Structure

SuperiorInsight/
|
|-- app.py                  # Flask application, RAG pipeline, API routes
|-- requirements.txt        # Python dependencies
|-- .env                    # Environment variables (not committed)
|
|-- faiss_index/            # Pre-built vector database (AI Brain)
|   |-- index.faiss
|   |-- index.pkl
|
|-- templates/
|   |-- index.html          # Chat interface
|
|-- static/
|   |-- style.css           # Frontend styles
|   |-- script.js           # Frontend logic
|
|-- README.md

---

Installation

Prerequisites

• Python 3.10 or higher
• pip
• A valid Groq API key

Steps

# Clone the repository
git clone https://github.com/your-username/superiorinsight.git
cd superiorinsight

# Create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

---

Configuration

Create a .env file in the root directory:

GROQ_API_KEY=your_groq_api_key_here

The FAISS index must be present in the faiss_index/ directory before running the application. If you are rebuilding the index from source documents, run the ingestion script separately.

---

Running the Application

python app.py

The application will be available at http://localhost:5000.

---

Deployment

Render

1. Push the repository to GitHub.
2. Create a new Web Service on Render.
3. Set the build command to pip install -r requirements.txt.
4. Set the start command to gunicorn app:app.
5. Add GROQ_API_KEY as an environment variable under the service settings.

Railway

1. Connect the GitHub repository to a new Railway project.
2. Add the GROQ_API_KEY environment variable.
3. Railway will auto-detect the Flask application and deploy.

Heroku

1. Ensure a Procfile is present: web: gunicorn app:app
2. Deploy via the Heroku CLI or GitHub integration.
3. Set the config var: heroku config:set GROQ_API_KEY=your_key

---

Dataset

The vector database was constructed from over 4,750 pages sourced from Superior University's official digital ecosystem, including:

• Undergraduate and postgraduate program catalogs
• Admission policy documents
• Fee structures and scholarship guidelines
• Campus facility information
• Faculty and departmental directories
• Academic calendar and examination regulations

All source material is institutional property of Superior University, Lahore.

---

Limitations

• Responses are bounded by the indexed knowledge base. Information not present in source documents will not be retrieved.
• The system does not browse the internet or access live university portals.
• The FAISS index must be manually rebuilt if source documents are updated.
• Multilingual translation accuracy depends on query clarity; heavily colloquial or ambiguous input may reduce retrieval precision.

---

Author

Syed Tahir BS Artificial Intelligence — 4th Semester Superior University, Lahore

---

SuperiorInsight is an academic project developed for educational purposes. It is not an official product of Superior University.