SDE2 @ Barclays — backend systems, distributed services, and AI agents. Active open-source contributor (SGLang, VS Code, Kubernetes).
- Shipping production backends and data-intensive systems at Barclays
- Merged into sgl-project/sglang — skip diffusion warmup preload when it would run out of GPU memory (#40761)
- Working on NVIDIA Dynamo — NIXL telemetry for disaggregated vLLM KV loads
- Working on VS Code chat — turn changes in the Agents window, on top of the Modern UI fixes already merged
- Building codereviewer_agent — multi-agent GitHub PR reviewer with unified RAG (code + JIRA + Confluence), security/pattern agents, and ensemble verification
| Area | What I actually ship |
|---|---|
| Backend | Java / Spring Boot, Go services, Kafka pipelines, Redis, PostgreSQL |
| AI / Agents | RAG (pgvector, ChromaDB), LangGraph, tool-calling agents, PII/compliance gateways |
| Distributed systems | Telemetry ingestion, billing/outbox patterns, notification fan-out, observability |
| Open source | LLM serving (SGLang, Dynamo), VS Code UX, Kubernetes, Playwright, Appwrite, Airflow |
- sglang#40761 — Skip diffusion warmup preferred preload when it would OOM
- kubernetes#140447 — Deprecate
endpoint_slice_controller_changesin favor of_totalsuffix (stable metrics) - vscode#331612 — Center Modern UI panel title tabs in the 32px header
- vscode#330754 — Modern UI tabs: reserve close-button column so it doesn't overlay filename
- vscode#325291 — Fix Modern UI editor tab decoration color on full label
- playwright#41845 — Pass client name on token-bypass connect
- appwrite#12891 — Return ISO8601 session expire from
updateSession - pkl#1383 — Fix super method call inside
letexpression - gradient-python#56 — Add
wait_until_ready()for agent deployment polling
- codereviewer_agent — Multi-agent PR review pipeline: Supabase/pgvector RAG, parallel security + pattern agents, LangGraph ensemble, GitHub Actions
- policyguard — Compliance RAG + PII redaction gateway (FastAPI, pgvector)
- a2a-incident-agents — Diagnosis + remediation agents in Spring Boot (A2A-style, approval gates)
- usage-billing-pipeline — Go billing pipeline: Kafka, Redis dedup, outbox, DLQ, Prometheus
- instrumentation-service — Go telemetry ingestion (gRPC/REST, Kafka, PostgreSQL)
- CopilotPulse — Local-first VS Code extension for Copilot Chat usage analytics
Marathons, weight training, and too many terminal tabs.