SDE2 @ Barclays · backend, distributed systems, AI agents, Inference
12 PRs merged upstream into vLLM, SGLang, NVIDIA Dynamo, Kubernetes and VS Code
Now
- Building codereviewer_agent: multi-agent PR reviewer with unified RAG
- Building briefly: AI briefing agent on Cloudflare Workers AI
- In review: sglang#43871 · dynamo#16108 · dynamo#16416 · sglang-omni#2842 · mcp-digitalocean#478 · vscode#328582
Recently merged
- vllm#62046: carry the boundary tie count across blocks in the split top-p mask kernel
- vllm#60885: reject mismatched float16/bfloat16 KV cache dtype
- dynamo#15570: keep NIXL telemetry capture when export is disabled
- sglang#40761: skip diffusion warmup preload when it would OOM
- kubernetes#140447:
_totalsuffix for endpoint slice controller metric - vscode#331612 · #330754 · #325291: Modern UI tab fixes
More
- playwright#41845: pass client name on token-bypass connect
- appwrite#12891: ISO8601 session expire from
updateSession - pkl#1383: super method call inside
let - gradient-python#56:
wait_until_ready()for agent deployments
Projects
- codereviewer_agent: LangGraph PR review pipeline with pgvector RAG
- policyguard: compliance RAG + PII redaction gateway
- a2a-incident-agents: diagnosis + remediation agents in Spring Boot
- usage-billing-pipeline: Go billing on Kafka, Redis, outbox, DLQ
- CopilotPulse: local-first Copilot Chat analytics for VS Code
Marathons · weight training · too many terminal tabs





