Quantitative Research using Snowflake, Machine Learning and Artificial Intelligence

Introduction, Set up, Workshop Overview

• • Goal: Outline workshop objectives, set expectations, introduce data sources and the Snowflake Financial Data Cloud + AI Platform

• Topics:
  • Purpose of combining structured and unstructured data
  • Overview of data sources (e.g., historical prices, earnings transcripts)
  • Brief intro to tools & libraries (Snowflake, SQL, Python, Snow Pandas, LLMs, vector embeddings, vector search)

Session 1 | Structured Data Analysis

• Goal: Demonstrate the tools available on Snowflake for importing, transforming, visualizing and analyzing structured data.

• • Topics:
    • Overview of structured data sources and how to access them on the marketplace
  • • Data extraction, cleaning, and preprocessing techniques
  • • Exploratory Data Analysis (EDA) on historical stock prices
  • • High performance time series analysis Snowflake 

• Hands-On Activity: Calculating basic indicators (e.g., moving averages) for a sample asset. Monte Carlo simulation and forecasting.

Coffee Break

Session 2 | Unstructured Data Analysis with LLMs (LLaMA, Mistral) and Vector Embeddings

• Goal: Introduce unstructured data analysis with a focus on using LLMs and vector search.

• Topics:
  • Overview of unstructured data sources (earnings transcripts, FOMC press conferences, filings)
• • Generating vector embeddings from text data
• • Fundamentals of vector embeddings and vector space representations
• • Vector search concepts: finding similar content, similarity scores, and their relevance in finance
• • How LLMs differ from traditional NLP models for financial text

• Hands-On Activity:
  • Extracting signal from text using Cortex, vector embeddings and foundational LLMs

Lunch

Session 3 | Merging Structured & Unstructured Data & Developing a Simple Trading Strategy

• Goal: Show how to combine insights from structured and unstructured data.
  

• Topics:
  • Data alignment techniques (e.g., matching timestamps, data transformations)
• • Feature engineering with mixed data types
• • Feature Store best practices
• • Simple strategy construction

• Hands-On Activity:
  • Merge historical price data with vectorized sentiment scores to create a combined dataset.
• • Feature engineering and storing

Q&A, Wrap-Up, and Next Steps

• Goal: Address questions, recap key takeaways, and provide resources for further study.

• Topics:
  • Summary of key concepts
• • Suggested resources and tools for continued learning
  • Open discussion on additional use cases and challenges