The Problem
Weekly production data work depended on manual sequencing across ingestion, compliance, prediction, and reporting stages. A missed step or partial failure could silently break downstream outputs, and non-technical operators had limited visibility into what happened.
Rui's Role
Rui designed the orchestration architecture, implemented the LangGraph state machine, connected stage-level execution and monitoring, and made failure diagnosis usable for operations without exposing sensitive data.
Engineering Decisions
- Gate-node routing: downstream work only runs when prerequisites succeed, reducing bad output propagation.
- Parallel fan-out: LRE, compliance, prediction, and LEA report stages can run concurrently after ingestion.
- Typed state reducer: concurrent node writes are merged predictably instead of mutating shared state ad hoc.
- Local LLM diagnostics: Ollama analyzes logs on the server, preserving privacy while helping operators recover faster.
- Rerun-safe execution: stage boundaries and logs make partial failures visible and restartable.
Capability Signal
This project shows Rui's ability to combine AI workflow design, backend orchestration, data platform reliability, privacy constraints, and operator UX into one production-minded system.
Where To Go Deeper
The homepage keeps this project short for recruiters. The Engineering Log preserves the complete technical details, stack, and operational impact.