Unlocking AI Innovation: How LLMs Work and Their Application in Finance

Join us for an evening of AI innovation, exploring how Large Language Models (LLMs) work and their applications in finance. Jonas Thunberg, Data Scientist at Solita, will break down the mechanics of LLMs, and Dmitrii Shiriaev, Machine Learning Engineer, will showcase how his team built a Retrieval-Augmented Generation (RAG) system for financial data in just 36 hours. Don’t miss this opportunity to learn and network. Food and drinks are sponsored by Solita.

Agenda:
17:30 - 18:00: Doors open
18:00 - 18:10: Welcome
18:10 - 18:40: How do LLMs work? Building intuition on Neural Networks, Transformers, and LLMs
18:40 - 19:10: Break
19:10 - 19:40: Building RAG for Financial Data: LLM-Powered Hackathon Success in 36 Hours
19:40 - 20:30: Networking

How do LLMs work? Building intuition on Neural Networks, Transformers, and LLMs
Jonas Thunberg - Data Scientist, Solita

Modern LLM:s are implementations of an artificial neural network model called a Transformer. During this presentation, I will try to give an intuition why these models are capable of generating text with such fidelity. My starting point will be networks consisting of a few artificial neurons, exploring hands on how these can be configured to perform simple tasks. I will describe word embeddings, the transformer architecture and end with a deep dive into how its' so called "attention blocks" learn to generate language. I aim to keep this talk accessible regardless of prior knowledge of machine learning, but I hope the intuitions we build will be rewarding (and useful) also to those more experienced.

Speakers Bio: Jonas Thunberg is a Data Scientist with over three years of experience, specializing in machine learning, data-driven product development, and customer behavior modeling. He is proficient in Python, SQL, as well as deep learning models, and is currently consulting at Solita while holding a Master’s degree in Machine Learning.

Building RAG for Financial Data: LLM-Powered Hackathon Success in 36 Hours
Dmitrii Shiriaev - Machine Learning Engineer

The challenge of extracting relevant information from financial reports, often stored in complex PDF formats, creates significant obstacles for both automated systems and human experts. These documents, filled with tables, specialized formatting, and scanned pages, hinder investors’ ability to quickly find key data, leading to delayed decision-making, particularly at scale.

To address this challenge, we developed an AI-driven dialogue-based interface during a hackathon, which allows users to interact directly with financial reports. By combining advanced AI techniques for text and table analysis, knowledge graphs for metadata, and LLMs with retrieval-augmented generation (RAG), our solution answers user queries in natural language while ensuring data provenance. This approach not only won the competition but also has the potential to significantly enhance the efficiency of financial report analysis for investors and analysts.

Speakers Bio: Dmitrii Shiriaev is a Machine Learning Engineer with interest in both classic ML and NLP fields. With expertise in Python, SQL, and automation, he is passionate about making data-informed decision process fast and accurate, transforming complex data into actionable insights.

–

About the event:
Tickets: Sign up required. Anyone who is not on the list will not get in. The event is free of charge.
Capacity: Space is limited. If you are signed up but unable to attend, please change your RSVP 2 days before the event.
Food and drinks: Food and drinks are sponsored by Solita.
Questions: Please contact the meetup organizers.

–

Code of Conduct

The NumFOCUS Code of Conduct applies to this event; please familiarize yourself with it before attending. If you have any questions or concerns regarding the Code of Conduct, please contact the organizers.