This video demonstrates how to use the Gemini API to extract and analyze financial predictions made by YouTube influencers and TV personalities. Learn to build a Python application that processes video content, extracts structured prediction data using Gemini's multimodal capabilities, and evaluates prediction accuracy against Yahoo Finance data. The tutorial covers setting up dependencies, comparing Gemini Pro vs 2.0 Flash models, creating structured extraction prompts with Pydantic models, and analyzing real predictions from financial commentators like Joseph Wang and Joseph Carlson. The walkthrough includes both successful and unsuccessful prediction examples, concluding with suggestions for extending the concept into a comprehensive prediction search engine. All source code is available on GitHub for those wanting to implement or expand upon this prediction analysis system.
Using Gemini API to Extract and Analyze YouTube Predictions
Overview
Syllabus
0:00 Intro
1:00 Project Description - processing predictions made in YouTube videos and what happened since
2:12 Begin code walkthrough, dependencies
5:41 Gemini client, Gemini Pro vs 2.0 Flash, now allows YouTube URLs as input
7:08 Structured extraction task, prompt for extracting predictions, pydantic models
8:51 Joseph Wang's excellent macro predictions and S&P 500 targets for 2024, 2025
11:33 Calling Gemini API in Python, Joseph Carlson's predictions
14:15 Predictions that didn't go so well
15:54 Wrapping up, ideas to extend this concept into a larger search engine
Taught by
Part Time Larry
