Context Engineering
The discipline behind great AI systems: deciding what goes into the model's context window — retrieval, memory, tool results, and compaction. Best if you're comfortable with Python or JavaScript and basic prompting.
Context windows & tokens
What actually fits in a model's context, what it costs, and how attention degrades.
Principles of context engineering
Treat context as a scarce resource: curation, ordering, and signal-to-noise.
Retrieval & RAG
Chunking, embeddings, hybrid search, and reranking to bring in the right knowledge.
Memory & state
Persist what matters across turns and sessions: summaries, scratchpads, and external memory.
Tool results & compaction
Feed agents tool output without drowning them; summarize and prune long histories.
Long-context evaluation
Measure retrieval quality and long-context degradation instead of guessing.