My Work Experience

Working on agentic-AI applications. An agent is a black box until you can see every prompt and every step it takes, so the first job was making it visible.

The numbers:

• Token usage down 5% after the traces showed where it went

What I did:

• Self-hosted LangFuse and OpenTelemetry so every prompt and workflow step is traceable
• Tune prompts and workflows against evidence from those traces instead of guesswork

Joined as Software Engineer II and was promoted to Senior Software Engineer in June 2025. Two systems: one about speed, one about taking a manual queue off people's desks.

The numbers:

• Orchestration latency from 20 seconds to under one, a 95% cut
• Policy issuance time down 72.3%
• Above & Beyond Award, 2025, for the agentic validator

What I did:

• Built a reactive orchestration engine on Java Easy Flows and MongoDB Change Streams, with virtual threads and a reactive WebClient
• Built a RAG-driven agentic validator that reads e-applications, finds the defects a reviewer would, and writes the fix-sheet itself

Started as Software Engineer and was promoted to Senior Software Engineer in April 2024. High-throughput, event-driven services that the rest of the network leans on.

The numbers:

• 2B+ shipment updates a day on the scan-push service
• 10× faster event processing on Waybill, and fault tolerance worth a 3× business gain
• 70% off billing execution time
• Above & Beyond Award, 2023, for leading TMS SaaS delivery
• Pinnacle Problem Solver, three times, 2022

What I did:

• Built the scan-push service: a Kafka factory architecture on multi-threading and AsyncIO
• Moved Waybill to multi-threading and Vert.x
• Rewrote billing on Spring and Spark with Drools rules cached in Redis
• Led the FastAPI and Kafka integration layer for a TMS SaaS MVP that onboarded six pilot clients

One of the core maintainers. I own the retrieval side and the release pipeline.

What it does:

• Own the RAG engine (LangChain, LangGraph, QdrantDB) that reverse-engineers how ATS software scores a resume
• Own the GitHub Actions and Docker pipeline that ships releases

Ask an agent what is paging you and who is on call, and let it act on the answer.

The numbers:

• 82 verified tools across alerts, schedules, teams, maintenance windows and routing
• 28 registered by default, so the agent starts narrow

What it does:

• Search alerts and read their notes and activity timeline
• Acknowledge, close, annotate and escalate alerts, and confirm the write actually landed instead of assuming it did
• Look up who is on call now and next
• Published on npm and listed on the MCP Registry and mcpservers.org