Why Search Is Broken & What Comes After RAG

Italian computer scientist based in Hong Kong, specializing in search, retrieval algorithms, and semantic tagging. Background in International Business (Tamkang University), quantitative finance, EY Milan, and AS Watson. Self-taught in CS during COVID, 100+ technical articles published on Medium (medium.com/@ardito.bryan), and author of the Midjourney-illustrated webcomic "The Leopard: Dawn of the Warrior" (webtoons.com). Currently working on weighted/covariate search algorithms that address the RAG ceiling.

Chapters

1. 0:00Intro — AI Tinkerers reunion & meeting Michelangiolo
2. 2:14Search & retrieval — Michelangiolo's obsession
3. 4:15The Excel analogy — filtering a million rows before the LLM
4. 6:49Semantic search vs keywords — lamp, light & 10 years of stagnation
5. 8:48Making traditional search smarter — the RAG ceiling
6. 11:05Art search — when a tiger painting is called "Blue Ocean"
7. 13:55Trees vs graphs — how humans classify knowledge
8. 17:05Abstraction & cognition — categories, autism & Chinese culture
9. 20:00Picasso across categories — the ontology problem
10. 21:30Finance to CS during COVID — self-taught in Florence
11. 24:02Why computer science is uniquely self-teachable
12. 27:00Tags vs keywords — Steam, retail & the labeling gap
13. 29:24Resume search — one of the hardest domains
14. 33:07Covariate search — weighted multi-dimensional queries
15. 36:45Algorithmic tagging — outperforming zero-shot LLMs at scale
16. 42:04Noise in text — why keyword extraction still breaks
17. 46:08The Crimson Duke — a 460-page Lovecraftian AI comic
18. 48:40Probabilistic consistency — the character generation trick
19. 51:02Creative process — planning 20-page chapters in Figma
20. 54:48Inpainting & prompt consistency — the fake characters problem
21. 60:46Outro — future collab teaser & subscribe

About This Episode

Michelangiolo quit quantitative finance during COVID to teach himself computer science from his home in Italy — now he builds some of the most interesting search infrastructure in Hong Kong. We dig into why vanilla RAG is broken, his "covariate search" approach that applies variable weights across multiple semantic dimensions at once, and an algorithmic tagging method he's publishing that aims to outperform zero-shot LLM labelling at scale. Then the fun part: Michelangiolo wrote and illustrated a 460-page Lovecraftian comic called "The Crimson Duke" using Midjourney — discovering a probabilistic consistency trick for character generation along the way.