Tech Lead · Software Architect · Agentic AI EngineerOpen to new work · İzmir, Turkey · UTC+3
Station01
Agent Studio at RecVue
A LangGraph multi-agent orchestration layer and its agent harness, built into an enterprise revenue and billing platform.
What Agent Studio does, drawn
Led the team that built an agentic-AI layer into an enterprise revenue and billing platform, and architected the orchestration system underneath it.
The orchestration layer is a graph set with a bounded agent harness around it: a ReAct loop that cannot run away, a human-in-the-loop gate on the decisions that warrant one, circuit breakers on the calls that leave the process, and orphan recovery for the runs that stop mid-flight. Evaluation is heuristic where a rule will do and model-judged where it will not, with budget guardrails on the spend and conversation memory checkpointed into Postgres.
Underneath it sits a ground-up rewrite. The legacy Oracle model was replatformed onto PostgreSQL with an outbox-driven core and a streaming edge for the agents, so the agent layer reads a committed event rather than a table it might be mid-write on.
● Evidence
Each figure is a résumé figure and says what it counts
10LangGraph graphs in the orchestration topology
43nodes across that graph set
390+tables in the PostgreSQL model of the ground-up rewrite
332HTTP API groups in the replatformed system
14schemas in that same model
6Grafana dashboards in the observability layer
10-engineerteam, led on the enterprise billing platform