Boost LLM Accuracy on Financial Docs from 10% to 30%+ with These Techniques

What if you could get large language models to accurately analyze complex financial documents like 10-Ks and earnings calls?

In this technical tour de force, AI startup founder Natan Vidra demonstrates how novel fine-tuning techniques can lift LLM performance on financial Q&A from 10% to 30%+ accuracy.

You'll learn battle-tested methods including:

  • Parameter-efficient LoRA/QLoRA fine-tuning
  • Retrieval augmented generation enhancements
  • Query expansion, re-ranking, metadata filtering

See real results applying these approaches to 10-Ks, analyst reports and more.

Vidra also delves into the cutting-edge challenge of evaluating LLMs without human labels using new tools like RAGAS and LLMEval.

Perfect for financial enterprises looking to supercharge document analysis to quickly answer critical questions, find comparable companies, and extract key insights.

Watch the full talk to learn how to boost LLM accuracy and save significant time and money.

Human-centric AI - Natan Vidra | Mindstone
IntroductionWell, thanks so much initially to the MindStone team for setting this up. Really appreciate Josh and this whole community building AI in New York an

Read more