Financial Analyst RAG Chatbot

Financial Analyst Chabot is a RAG based solution that allows the user to ask questions based on financial documents (excel in this case). It provides a web based UI (Streamlit) for user interaction and uses OpenAI to generate embeddings and responses. PineCone is used to store and search vector embeddings.

Features

• Upload & parse excel spreadsheets
• Smart document chunking for optimal retrieval
• Created embeddings using OpenAI's text_embedding-3-large
• Vector store and similarity search using PineCone
• Chatbot interface built using Streamlit

See the demo below

Query 1 "What was the revenue last year?"

Query 2 "What are the top 3 liabilities?"

Query 3 "give me EBIT"

Tech Stack

• Embeddings - Open AI text-embedding-3-large
• LLM - Open AI gpt-4o
• Vector Store - Pinecone
• UI - Streamlit
• File Parsing - pandas + OpenPyXL
• Chunking - LangChain

Flow

First the documents are ingested using a parser and chunks are created using RecursiveCharacterTextSplitter from Langchain. Then using OpenAIEmbeddings text-embeddings-3-large embeddings are created and stored in the Pinecone vector database.

When a user asks a questions, chatbot create query embeddings and retrieves for similar documents using cosine similarity. Top_k is set to 8 and OpenAI GPT temperature is set to 0.2 to control the randomness & get realiable answers as we're dealing with financial documents.

📁 Project Structure

financial-rag-chatbot/
│
├── data/RAWDATA/               # Raw Excel files
├── src/
│   ├── ingestion/              # File parsing, chunking
│   ├── embeddings/             # Embedding generation
│   ├── vector_store/           # Pinecone connection and upsert
│   ├── chat/                   # RAG: retriever, prompt, LLM
│   ├── ui/                     # Streamlit chatbot
│   └── utils/                  # Config loader, token counter
├── config.yaml                 # API keys and config
├── requirements.txt            # Dependencies
├── main.py                     # Indexing pipeline
├── README.md                   # Documentation

Setup Instructions

1. Clone the repo

git clone https://github.com/prashantcodelab/financial-rag-chatbot.git
cd financial-rag-chatbot

2. Create a virtual environment and install dependencies

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

3. Configure your API keys in config.yaml

open_ai: sk-<your-openai-key>
pinecone: <your-pinecone-api-key>
pinecone_index: financial-rag-index

4. Ingest and Index Excel File

• Sample excel file is provided. You can use your own excel file and keep it in data/RAWDATA/

python main.py

This step will parse and chunk excel, embed the content and add vectore to Pinecone

5. Launch the Chatbot

streamlit run src/ui/streamlit_app.py

You can ask questions based on your excel file - "What is the total expense ?", "What are the top 5 vendors by payment amount?"

Future Updates

• PDF & CSV support
• Allow file upload directly from chatbot UI
• Role-based chat access
• Latency improvements - Caching

Additional details about Ingestion and Chatbot

Ingestion - main.py

python main.py

• main.py - Runs the document ingestion pipeline. After first run, run only when new files are added.
  • Loads config -> config_loader.py
  • Parses file -> file_parser.py
  • Document chunking -> chunking.py
  • Embed chunks -> embedder.py
  • PineCone -> pinecone_client.py
  • Upserts vectors to Pinecone

main.py
├──→ config_loader.py      # loads API keys
├──→ file_parser.py        # reads Excel files
├──→ chunking.py           # splits into text chunks
├──→ embedder.py           # embeds chunks into vectors
└──→ pinecone_client.py    # connects and upserts to Pinecone

Chatbot - streamlit_app.py

streamlit run src/ui/streamlit_app.py

• streamlist_app.py - Chatbot
  • Loads config - config_loader.py
  • Pinecone - pinecone_client.py
  • Query embedding - embedder.py
  • Retrieves context - context_retriever.py
  • Builds prompt - prompt_builder.py
  • LLM invocation - llm_interface.py
  • Responses in chatbot UI

streamlit_app.py
├──→ config_loader.py         # loads API keys
├──→ pinecone_client.py       # connects to Pinecone
├──→ embedder.py              # embeds the user's query
├──→ context_retriever.py     # finds top-k matching chunks
├──→ prompt_builder.py        # builds final LLM prompt
└──→ llm_interface.py         # gets response from OpenAI

Source for Sample Data File - https://www.nasdaq.com/market-activity/stocks/saia/sec-filings