Integrating LLMs with Traditional Data Analysis
summary
Shows concrete integrations and results: LLMs as feature extractors feeding an SQL-backed ML pipeline, synthetic data generation to fix class imbalance, and statistical validation of LLM outputs using ensemble and grounding with vector DB searches. Reports a production hybrid sentiment system with side-by-side metrics (traditional-only, LLM-only, hybrid) showing hybrid wins on accuracy and interpretability while lowering compute vs LLM-only; surprising wins include random forest on 80 labeled examples and ML beating LLMs on a 300–400 diary study.
AI-written summary · about
chapters (9)
- 0:00LLM and analytics integration
- 1:04LLMs need classical ML
- 3:53LLM-ML integration patterns
- 6:10AI matchmaking case study
- 9:25Predicting viral content
- 13:19Automating model development
- 16:09Explaining ML results
- 19:26Validating data readiness
- 21:18Q&A
related talks
33mDesigning LLM Applications for Massive Datasets
30mWhat Can LLMs Do with Messy Data?
15mLLMs for Data-Driven Decision Making in Product Development
28mExplaining Machine Learning Models with LLMs
49mA Practical Approach to Using LLMs: Peculiarities and Pitfalls
47mGive Your LLMs a Left Brain
35mEvaluating LLM systems: the nuts and bolts
17mFrom messy to meaningful data: LLM-powered classification in R