Engineering notes
How we build agents, retrieval and ML systems that have to work in production. Most posts include the real code and screenshots behind them.
Latest · AgentsPausing an agent for a human: LangGraph interrupts and checkpoints
Make an agent stop before an irreversible action and wait for a person, with LangGraph's interrupt and a checkpointer. Graph, code and real output.
14 Oct 2026 | 5 min read | Vithupro EngineeringRead the article
OrchestrationAgents that can wait three days: a durable refund workflow on Temporal
A refund agent that waits days for human approval and survives restarts. The workflow, the code and what Temporal's UI shows, from a run on our machines.
14 Oct 2026 | 6 min read
MLOpsFrom four runs to a champion model: experiment tracking with MLflow
Tracking a forecasting experiment in MLflow: four runs, metrics side by side, and a registry alias that serving reads. Real screenshots from our run.
14 Oct 2026 | 5 min read
RetrievalWhy enterprise RAG hallucinates, and the four fixes that stop it
Naive RAG still hallucinates. Four techniques (hybrid search, re-ranking, GraphRAG and grounding guardrails) that make enterprise RAG accurate and cited.
10 Jul 2026 | 8 min read
AgentsGiving agents tools without giving them the keys
Giving AI agents real tool access safely: least-privilege scoping, MCP as the boundary, human-in-the-loop approvals and auditing every call.
10 Jul 2026 | 7 min read
EvaluationYou can't ship what you can't measure: evals for LLMs
Why traditional tests fail for LLMs, and how to build an evaluation harness that scores grounding, accuracy and safety on every release.
10 Jul 2026 | 7 min readHave an AI system to build?
Tell us what it needs to do. We reply within one business day with questions, a rough plan and the next step.
- Free 30-minute architecture call
- NDA before you share details
- You own all code and IP
- Reply within one business day