Transforming Education: Applying AI to Universities
summary
Instructure embeds course content with Cohere embeddings and stores vectors using PGVector in Postgres, then quantizes them to cut storage from about 5 TB to 1.3 TB (8-bit ints, ~96% search accuracy) or 164 GB for binary (≈90% accuracy); an agent pipeline uses Anthropic Claude for planning, classifies whether a tool is needed, maps plans to API calls, and adds confirmation for destructive writes; adding all 800 endpoints inflates latency from ~3s to ~21s and token cost to ~0.50 per request.
AI-written summary · about
chapters (10)
- 0:00AI in education
- 2:26Teachers post-pandemic
- 7:03Trust in AI for educators
- 10:16AI hallucinations risks
- 13:43Embeddings for AI search
- 18:37AI automation for teachers
- 21:52Agentic workflows challenges
- 28:18AI student outcomes
- 36:46Decentralized education future
- 37:28Q&A on AI models
related talks
44mIntegrating AI with Postgres...
19mPOC to PROD: Hard Lessons from 200+ Enterprise GenAI Deployments
27mPutting AI in Production
18mAccelerate generative AI development with integrated data platforms
46m"Embeddings - the lesser known Hero of AI"
16mEmpower AI: Building Smarter Solutions with Embeddings
18m: Workforce Efficiency with AI-Driven Vector Search
32mAI in Education | How AI Is Transforming Learning