Overview #
I was invited by my friend Glenn to his company to give a workshop about AI and how it can influence financial services work. I really enjoyed talking about the topic and had a very nice discussion afterwards.
This workshop should explore the transformative potential of artificial intelligence in financial services, featuring a comprehensive discussion on AI’s evolution and practical applications. I started the session with a historical perspective on AI development, emphasizing that the current AI revolution is primarily driven by Large Language Models (LLMs) and a select few other model architectures. Through live demonstrations using ChatGPT, I showed various AI use cases that are reshaping how financial institutions operate, from customer service automation to complex data analysis and decision-making support.
Key Demonstration: Invoice Processing Architecture #
The centerpiece of the workshop was a hands-on demonstration of a local LLM system designed for automated invoice processing. I built the application layer inside python code and it used a local instance ollama to serve the 8B LLAMA LLM with tool calling.
Architecture Diagram #
The architecture showcased a sophisticated pipeline where PDF invoices are processed through a local language model, which extracts critical information including invoice details, dates, and billable amounts. The system then intelligently maps this extracted data to appropriate accounting categories, automatically assigning items to different accounts (such as Account A for VAT-related entries and Account B for standard billing amounts) before routing everything through a human validation step. This demonstration illustrated how AI can streamline financial workflows while maintaining necessary oversight and accuracy controls, representing a practical application of how modern LLM technology can enhance traditional accounting processes.
Slides #
