The Fifth Elephant 2023 Monsoon

Abstract

Auquan is an AI startup that serves institutional investors and investment managers with curated news and documents to help them make better investment decisions.

In this presentation, I will discuss our approach for using a vector database and a tuned language model to curate news items at scale.

I will walk through the process and pitfalls for using these technologies in production, and provide best practices for achieving high performance. Specifically, I will discuss the metrics that we used for selecting a vector database and tune our language model.

Audience

ML Engineers, early stage Data Scientists

Takeaways

• How to use a vector database and a language model to curate news items at scale
• Best practices for using vector databases in production
• Pitfalls to avoid when using vector databases

Presentation Outline:

• Introduction
  • About Auquan
  • Problem description
• Vector databases for news curation
  • Choosing a vector database
  • Using embeddings with vector databases for different tasks
  • Offline population and real time inference
• Tuning a language model for news curation
  • Tuning an LM
  • Using tuned model for embedding
• Stack architecture
• Conclusion/QA

Link to the Presentation

https://drive.google.com/file/d/1FubDUOYBJhzgaW-H6qxrlkEO49d3abDr/view?usp=sharing