Modern Problems Require Modern Solutions
summary
A Dockerized demo embeds 72 meme images with a Hugging Face image-feature-extraction model (z_nova/clip_vit_base_patch32) into 512-dim float32 embeddings, stores them in Redis as a vector DB, and compares cosine, L2 (Euclidean) and inner-product searches using flat and HNSW indices. It runs locally on an Intel MacBook, supports semantic caching and RAG, and the webcam-based “meme twin” lookup works in the live demo.
AI-written summary · about
chapters (11)
- 0:00Introduction to talk
- 1:03Who is the speaker
- 2:05Talk agenda overview
- 4:12What are embeddings
- 15:29Vector database indices
- 25:29How neural networks work
- 34:36Should you build models
- 36:23Demo setup
- 38:50Code walkthrough
- 44:30Finding your meme twin
- 48:25Resources and Q&A
related talks
59mC++ in containers
39mApps, Algorithms & Abstractions
43mAI w/o Spaghetti: Clean Architecture in the Age of AI
01h 00mThe Salami Method for Cross Platform Development
01h 01mArchitecture Decisions: A Case Study -
01h 00mNext-Gen Software Architecture with Large Language Models (LLMs) like ChatGPT
41mSRE & Complexification: Where Verbs and Nouns Do Battle
58mPatterns, Predictions and Programming