Now · Jun 2026 – Present
Tech Lead — Synup
I lead the AI product line — architecture, retrieval, tool surface, evals and inference, and the team that ships it. I also build the internal platforms the engineering org runs on.
Architecture
LangGraph state machines — tool use that is explicit and resumable.
Retrieval
RAG over a SQL vector store, grounding measured not assumed.
Tool surface
MCP servers built and integrated; the tool catalogue re-architected.
Observability
Every run traced in LangSmith, so regressions surface early.
Evaluation
Golden conversations gating every prompt and model change.
Inference
Open-weight models self-hosted on Runpod, chosen per workload.
80%
fewer tokens per request
6
stages owned, architecture to inference
- Took an agent from research question to production, then built the eval harness before scaling it — nothing ships unless the run is green.
- Cut tokens per request by 80% by re-architecting how the tool catalogue reaches the model.
- Replaced a third-party issue tracker with an in-house platform and drove the migration — pilot team, automated import, parallel run, hard cut-over. An MCP integration was the lever: it offered something the incumbent could not, so the new tool became the path of least resistance rather than a mandate.
- Built that platform end to end on React, FastAPI, PostgreSQL, Redis, Kafka and Elasticsearch.
- Own the Python services behind the AI features alongside the React front end, and review on both sides of the stack.
- Mentored four junior engineers into full-time roles.
- LangGraph
- LangChain
- LangSmith
- MCP
- FastAPI
- SQL vector store
- Runpod
- React
Also worked with — Azure AI Foundry