SME Finance AI Agent

An AI-driven chatbot built with Flask and LangChain to provide Colombian SMEs with practical financial guidance. It ingests live government-published documents (DIAN, MinCIT, CCB), chunks them, embeds with OpenAI, and retrieves threshold-filtered answers on cash flow, taxes, financing and more.

Features

• Live Document Ingestion
  Automatically scrapes and parses official pages to keep regulations and tax codes up to date.
• Semantic Retrieval
  Uses OpenAI embeddings stored in ChromaDB and a CustomThresholdRetriever to surface only highly relevant chunks.
• Structured Advisory Protocol
  Enforces multi-section responses (intro, recommended steps, risks/points of attention, optional timeline, closing) via a tailored prompt template.
• Interactive Web UI
  Responsive Flask interface with session memory, quick-reply buttons, dark-mode toggle and typing indicator for a smooth user experience.

Tech Stack

• Language: Python 3.10+
• Web: Flask, Jinja2 templates, Requests, BeautifulSoup
• AI/ML: LangChain, ChromaDB, OpenAI Embeddings, ConversationalRetrievalChain
• Configuration: dotenv for environment variables

Configuration & Usage

1. Define environment variables (e.g. OPENAI_API_KEY) in a .env file or your shell.
2. Launch the Flask app and navigate to http://localhost:5000.
3. Ask any finance-related question for SMEs in Colombia and receive structured, actionable advice.

API Reference

Endpoint	Method	Description
/	GET	Load the chat interface
/	POST	Submit user query, return AI-generated reply

Next Steps

• Schedule automated re-scraping and vectorstore refresh to capture regulatory updates.
• Extend support to regional tax regimes, export compliance and payroll modules.
• Add user authentication and personalized dashboards for multi-user scenarios.

Contributing

Contributions are welcome! Please open issues or submit pull requests at
https://github.com/pablo-reyes8

License

This project is licensed under the Apache License 2.0.