Building an Agent from First Principles
Build an AI agent from scratch in Python: tools, the observe-decide-act loop, state, planning, error handling, termination and verification.
Blog
Learn with me
14 posts, newest first
October 2026
Context Engineering Is the New Prompt Engineering
Beyond better prompts: how to design, assemble, optimize and manage the information an LLM needs to solve a task.
What Actually Happens Between Your Prompt and the Next Token?
A deep technical guide to LLM internals, from raw text to token generation, with Python implementations from scratch.
The AI Caching Playbook, Part 10: The Complete Production Architecture
Bringing it all together: how to design caching as a first-class part of an AI system, from keys and versions to reliability, cost and rollout.
The AI Caching Playbook, Part 9: Cache Observability and Cost
Measure what the cache actually saves: latency, tokens, calls avoided, cost, ROI, hit quality and freshness, not just the hit rate.
The AI Caching Playbook, Part 8: Cache Invalidation and Correctness
Fast is useless if the answer is wrong: TTLs, versioned keys, dependency graphs, events, concurrency and testing for correct AI caches.
September 2026
The AI Caching Playbook, Part 7: Distributed AI Caching
Scaling AI caches: Redis Cluster sharding, hot keys, L1/L2 caches, stampedes, invalidation at scale, multi-region, failure handling and cost.
The AI Caching Playbook, Part 6: LLM Inference Caching
KV cache, prefix cache and prompt caching inside LLM inference: how they differ, what they cost in GPU memory, and how to design prompts for reuse.
The AI Caching Playbook, Part 5: Cache the Agent's Work, Not Just the Answer
Agent state, plans, tool and API results, workflow steps, checkpoints, idempotent writes and scoped sharing: caching the work an agent does.
The AI Caching Playbook, Part 4: RAG Caching
Make retrieval faster, cheaper and smarter: caching queries, embeddings, retrieval, reranking, chunks, context and answers without serving stale data.