Field notes from the work.
Practical writing on legacy exits, cloud migration, sovereign ML, and building agents you can trust with real actions.
Agents that plan, act, and terminate
How to build an autonomous planning agent that decomposes goals into subtasks, calls tools in a reasoning loop, manages its own state, and knows when to stop — with step limits, deduplication, and failure recovery built into the architecture.
.svg)
Building an AI invoice processing agent with structured outputs
A complete walkthrough of the structured output agent architecture input ingestion across PDF, image, and text formats, schema-driven LLM extraction, multi-layer validation for type safety and arithmetic consistency, retry logic with targeted feedback, confidence scoring, and API design for invoice processing at scale.
A practical guide to RAG: search, embeddings, and vector stores
A field guide to the decisions that shape a real retrieval-augmented generation system — from dense, sparse, and hybrid search, to why the embedding model you store with must be the one you retrieve with.