Engineering notes.
Notes on how we build production AI agents, data platforms, integrations and workflow automation. Each one covers a method we use, with code where it helps, and the mistakes we see most often.
All notes · 4
Topic · published · length- Production AI agentsSheet N-11
Voice agent architecture: latency, interruptions and state
An engineering note on voice agents: a latency budget for each stage, streaming text to speech in whole phrases, handling interruptions and keeping call state.
Updated Oct 20266 min read - Data platformSheet N-12
Revenue attribution as a reconciliation problem
Linking a customer's first contact to billing, across systems that share no key, is a record-linking job first. How we model it, match it and check it.
Updated Oct 20266 min read - Production AI agentsSheet N-13
Building RAG systems that hold up in production
How we build retrieval-augmented generation: chunking by structure, hybrid search, reranking, a context budget, checked citations and when to skip RAG.
Updated Oct 20267 min read - Workflow automationSheet N-14
The operations automation playbook
A method for automating operations work: measure where it waits, sort each step into rule, judgment or write, and keep a person on writes to systems of record.
Updated Oct 20267 min read