Building a Retrieval-Augmented Generation (RAG) Assistant with LangChain and FAISS

🧠 Abstract

This project presents a Retrieval-Augmented Generation (RAG) Assistant that leverages LangChain, FAISS, and Gemini 2.5 Flash to answer user queries using a custom Wikipedia-based knowledge base.
The assistant dynamically retrieves relevant information and generates context-rich, accurate responses.
Key features include session-based memory, dynamic knowledge base initialization, robust error handling, and integrated logging.

This publication outlines the assistant’s purpose, technical architecture, and potential use cases in education, research, and professional settings.

---

1. 🎯 What Is This About? (Purpose)

The RAG Assistant is designed to provide accurate, context-aware responses by combining retrieval and generation techniques.
It grounds its answers in verified data using content retrieved from Wikipedia, making it ideal for:

• Educational platforms
• Research assistants
• Knowledge-based automation systems
• Decision-support tools

The system ensures each answer is contextually relevant and factually supported.

---

2. 💡 Why Does It Matter? (Value & Impact)

Traditional AI assistants often produce hallucinated or generic responses.
This project mitigates that issue by retrieving trusted content before generating an answer.

Key Advantages

• Grounded Responses: Ensures factual accuracy through retrieval-based generation.
• Enhanced UX: Introduces the knowledge base dynamically at runtime.
• Session Continuity: Maintains context across multi-turn conversations.

Overall, this RAG system bridges the gap between retrieval precision and generative fluency, making it a valuable learning and research companion.

---

3. 🧩 Can I Trust It? (Technical Quality)

The assistant is built using a robust, modular tech stack:

Reliability Features

• Dynamic Prompt Formatting: Adapts the prompt based on the topic.
• Session Memory: Enables follow-up questions and continuity.
• Error Handling & Logging: Increases robustness and transparency.

---

4. ⚙️ Can I Use It? (Documentation)

Yes — the assistant is easy to set up and customize.

🛠 Key Features

• Dynamic Knowledge Base: Change the topic by updating a single configuration file.
• Session Memory: Keeps the conversation context consistent.
• Error Handling & Logging: Provides traceability for debugging and evaluation.

🚀 Setup Overview

# Clone the repository
git clone https://github.com/Fraol-D/RAG-Chatbot.git
cd RAG-Chatbot

# Install dependencies
pip install -r requirements.txt

# Run the app
python main.py


## 📘 More Documentation
See detailed setup steps, examples, and troubleshooting in the [GitHub repository](https://github.com/Fraol-D/RAG-Chatbot).

---

## 🧰 Technical Implementation

```python
# from knowledge_base import create_vector_store
from pipeline import create_rag_chain
import config

# Step 1: Create the knowledge base
vectorstore, _ = create_vector_store(
    topic=config.WIKIPEDIA_TOPIC,
    chunk_size=config.CHUNK_SIZE,
    chunk_overlap=config.CHUNK_OVERLAP
)

if vectorstore:
    # Step 2: Build the RAG chain
    prompt_template = config.PROMPT_TEMPLATE.format(topic=config.WIKIPEDIA_TOPIC)
    rag_chain = create_rag_chain(vectorstore, prompt_template)

    # Step 3: Run a query
    question = "What are the achievements of Haile Gebrselassie?"
    response = rag_chain.invoke({"question": question})
    print("Response:", response)
else:
    print("Failed to initialize the knowledge base.")

Knowledge Base Creation

• WikipediaLoader ingests topic content.
• RecursiveCharacterTextSplitter divides content into chunks.
• FAISS indexes chunks for efficient semantic retrieval.

Prompt Engineering

A dynamic prompt template tailors responses to the current topic, ensuring contextual precision.

Response Generation

• LangChain manages the overall RAG pipeline.
• Gemini 2.5 Flash produces rich, contextually grounded outputs.

Enhancements

• Multi-turn session memory for conversational flow.
• Error handling & logging for reliability and transparency.

---

🔍 Limitations and Future Directions

Current Limitations

• Dependent on the quality of the selected Wikipedia page.
• No real-time updates to the knowledge base.

Future Improvements

• Multi-knowledge-base integration.
• Live data updates for dynamic knowledge refresh.
• Multilingual support for global accessibility.

---

📬 Contact Information

Author: Fraol Ashebir Demisse
Email: fraolashebir84@gmail.com
GitHub: https://github.com/Fraol-D
Project Repository: https://github.com/Fraol-D/RAG-Chatbot

---

🖼️ Optional Enhancements

• Add architecture diagrams (e.g., RAG pipeline with LangChain + FAISS).
• Include screenshots of console output or chatbot interface.
• Embed snippets showing LangChain pipeline configuration.

---

Tags: LangChain FAISS RAG Gemini 2.5 Flash AI Assistant Python Machine Learning