Context-Aware Retrieval-Augmented Knowledge Inference Engine.

A modern Retrieval-Augmented Generation (RAG) application built using Google Gemini, LangChain, ChromaDB, and Gradio.

The system answers user questions strictly based on locally provided documents. It is designed to overcome fundamental limitations of Large Language Models (LLMs), specifically hallucination and lack of updated information beyond their knowledge cutoff.

---

🚨 Core Problems with Large Language Models

Problem 1: Hallucination

Large Language Models are probabilistic text generators. As a result, they often:

• Produce confident but factually incorrect responses
• Fabricate information when relevant knowledge is missing
• Do not clearly state uncertainty

This behavior makes LLMs unsuitable for enterprise, research, and decision‑critical applications.

---

Problem 2: Static and Non‑Contextual Knowledge

Large Language Models:

• Are trained on static, historical datasets
• Cannot access private, internal, or domain‑specific documents
• Cannot adapt to new or frequently updated information

This severely limits their usability in real‑world systems that require dynamic and context‑aware intelligence.

---

✅ How This Project Solves These Problems

• Uses relevant document content before answering, so the model does not guess
• Fetches fresh information from local files instead of relying on old training data
• Clearly responds with “I don’t know” when the answer is not found in documents
• Breaks documents into smaller chunks for faster and more accurate retrieval
• Separates ingestion, retrieval, and inference for easy updates and scalability

---

🧠 How It Works (RAG Flow)

1. Documents are loaded from the data/ folder
2. Documents are split into small chunks
3. Each chunk is converted into vector embeddings
4. Embeddings are stored in ChromaDB
5. User queries retrieve the most relevant chunks
6. Gemini generates answers using retrieved context only

---

🛠️ Tech Stack Used

• Programming Language: Python
• Large Language Model: Google Gemini
• Framework: LangChain
• Vector Database: ChromaDB
• Embeddings: Google Generative AI Embeddings
• Frontend / UI: Gradio
• Environment Management: Python venv, dotenv

---

⚙️ Setup Instructions

1️⃣ Clone the Repository

git clone https://github.com/supriya46788/Context-Aware-Retrieval-Augmented-Knowledge-Inference-Engine.git
cd Context-Aware-Retrieval-Augmented-Knowledge-Inference-Engine

2️⃣ Create Virtual Environment

python -m venv venv
venv\Scripts\activate

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Configure Environment Variables

Create a .env file in the project root:

GOOGLE_API_KEY=your_google_gemini_api_key

5️⃣ Ingest Documents

python ingest.py

Any update to the data/ directory requires re‑running the ingestion step.

6️⃣ Run the Application

python app.py

Access the application at:

http://localhost:7860