Searching for My Next Chart
summary
Builds a RAG-based retrieval system that turns scattered ggplot2 and Datawrapper visuals into a semantically searchable corpus, embedding plot code and VLM-generated descriptions and storing them in a vector database. Uses QN VLM to deconstruct images, Kim K2 for tool-calling, Mistral embeddings, Chainlit frontend and an LLM re-ranker; retrieves up to 300 (often 150) candidates then reranks. Claims cut curation time that used to be 85% of workflow; demo works but retrieval quality and evaluation remain subjective.
AI-written summary · about
chapters (22)
- 0:00Introduction
- 0:25Excel to ggplot2 journey
- 3:13Data visualization journey
- 5:02Tidy Tuesday & chart challenge
- 6:31First chart submissions
- 8:02Soviet visuals & screen printing
- 9:07Why creative visualization
- 10:10Dutch loan words
- 12:18Visualizing Gaza & buildings
- 13:25Chart collection inefficiencies
- 14:53AI for chart search
- 16:37Chart search demo
- 17:46How chart search works
- 19:16Embedding data visuals
- 21:26Grammar of graphics
- 22:33Deconstructing visuals
- 24:59Improving retrieval
- 26:46Longer context mode
- 28:53Challenges & evaluation
- 29:15Future improvements
- 29:37Perfectionism & next steps
- 30:41Q&A
related talks
33mVector Databases in the Age of AI: Embracing RAG for Enhanced Search Capabilities
30mGraph RAG: Bringing together graph and vector search to empower retrieval
01h 35mTutorial: Build AI-Supercharged RAG Apps with a Vector Database
49mPower up LLMs to be Factual Ninjas using GraphRAG
46mExplorations in Interactive Visual Analytics
31mRetrieval Augmented Generation
15mEvaluating RAG Model Performance Metrics, Bias, and Interpretability
44mRAG for a Medical Company: Technical and Product Challenges AI & Data