Real-Time Context Engineering for LLMs
summary
Includes a working TikTok clone wired to a Hopsworks feature store and application IDs that unlock per-application context for agents, showing a concrete pipeline: events → streaming features → feature store → MCP query → prompt assembly → LLM. Claims real-time freshness at human‑perceivable latencies (about one to two seconds), compares streaming versus batch pipelines, shows how structured features and embeddings are stored and retrieved, and gives numbers and architecture rather than vague examples.
AI-written summary · about
chapters (9)
- 0:00Real-time context engineering intro
- 3:21In-context learning for LLMs
- 6:15Agents and context injection
- 9:04Structured vs unstructured data
- 13:20Task decomposition for context
- 17:17Streaming pipelines for real-time
- 20:52TikTok-like recommendation agent
- 22:42Introducing Brewer
- 23:44Closing & Q&A
related talks
40mReal-TIme Context Engineering for Agents
37mContext is King in AI Agents
31mBuild Intelligent AI Agents With Context Engineering
27mContext is the new code: The context development lifecycle- 27mContext is the new code: The context development lifecycle
23mConnecting the Dots With Context Graphs
43mLLMs in the Wild - Streaming, RAG, and Real Time GenAI at Scale
38mQuerying Live Data With LLM App