## Project Overview
This respository was created as part for an internship report, containing the source codes and respurces in developing the Retrieval-Augmented Generation(RAG) banking chatbot.  

RAG allows the chatbot to answer questions based on the knowledge base instead of only using the Large Language Model (LLM). 

## Objectives
- Develop a chatbot capable of answering questions based on public dataset(s)
- Implement a Retrieval-Augmented Generation (RAG) pipeline
- Generate embeddings for efficient document retrieval
- Build a vector database
- Improve response accuracy by providing relevant context to the language model
- Implement advanced requirements
- Evaluate the chatbot's performance 

## Installation
Virtual environment:  
python -m venv venv  
venv\Scripts\activate  

Libraries:  
pip install -r requirements.txt  

Run the chatbot:  
streamlit run app.py  

## How it works
The chatbot follows a Retrieval-Augmented Generation (RAG) workflow:

1. Documents are collected and preprocessed.
2. The documents are split into smaller text chunks.
3. Each chunk is converted into vector embeddings.
4. The embeddings are stored in a vector database.
5. When a user submits a query:
   - The query is embedded into a vector.
   - The vector database retrieves the most relevant document chunks.
   - The retrieved context is combined with the user's query.
   - The LLM generates an appropriate response using the retrieved information.

## Requirements
- Python 3.10 or above
- LangChain
- Ollama
- Vector database (ChromaDB)
- Sentence Transformer / Embedding model
- python-dotenv
- Other dependencies listed in `requirements.txt`

---
license: mit
Datasets: [Bank Faqs](https://www.kaggle.com/datasets/somanathkshirasagar/bankfaqs), [customer support conversations](https://www.kaggle.com/datasets/syncoraai/customer-support-conversations)
---