SDE2 @ Barclays · backend, distributed systems, AI agents, Inference
19 PRs merged upstream into vLLM, vLLM-Omni, SGLang, NVIDIA Dynamo, Kubernetes and VS Code
About me
Akshat Anand is a Software Development Engineer II (SDE2) at Barclays, based in India, working on backend systems, distributed systems, LLM inference and AI agents. He has 19 pull requests merged into open-source projects including vLLM, SGLang, NVIDIA Dynamo, Kubernetes and VS Code.
name: Akshat Anand
title: Software Development Engineer II (SDE2)
company: Barclays
location: India
roles:
- Backend Engineer
- Distributed Systems Engineer
- LLM Inference Engineer
- AI Agents Engineer
certifications:
- AWS Certified Cloud Practitioner: https://www.credly.com/badges/d2a21830-7ca1-465c-bd95-3eb4e5187224/public_url
- AWS Certified AI Practitioner: https://www.credly.com/badges/b7864762-eaf0-4cd8-8079-d89291e7159c/public_url
skills:
languages: [Java, Go, Python, TypeScript]
backend: [Spring Boot, FastAPI, REST APIs, gRPC, Microservices]
data: [Apache Kafka, PostgreSQL, Redis, pgvector]
distributed_systems:
- Event-driven architecture
- Transactional outbox
- Idempotency
- Prometheus observability
llm_inference: [vLLM, SGLang, NVIDIA Dynamo, KV cache, Sampling kernels]
ai_agents: [RAG, LangGraph, Tool calling, Multi-agent systems]
cloud_devops: [AWS, Docker, Kubernetes, GitHub Actions, Cloudflare Workers]
open_source:
merged_prs: 19
projects:
- vLLM
- vLLM-Omni
- SGLang
- NVIDIA Dynamo
- Kubernetes
- VS Code
- Playwright
- Appwrite
- Apple PklRecent contributions · 2026
- ✅ NVIDIA Dynamo: wait for the KAI queue controller before creating Dynamo queues in the Helm chart (#16723)
- ✅ VS Code: release
code -w -g file:lineonce the file's editor closes (#344740) - ✅ NVIDIA Dynamo: end-to-end AICR installation guide for Kubernetes (#16416)
- ✅ vLLM-Omni: place stage-CLI test replicas on separate GPUs (#9041)
- ✅ VS Code: keep the paused debug location range valid in chat (#342379)
- ✅ vLLM: serve Gemma 4 on
/v1/systemoneand/v1/decisions(#62303) - ✅ SGLang: clamp glm4v request video budgets to the deployment limit (#43871)
- ✅ vLLM: carry the boundary tie count across blocks in the split top-p mask kernel (#62046)
- ✅ vLLM: reject mismatched float16/bfloat16 KV cache dtype (#60885)
- ✅ NVIDIA Dynamo: keep NIXL telemetry capture when export is disabled (#15570)
- ✅ SGLang: skip diffusion warmup preload when it would OOM (#40761)
- ✅ Kubernetes:
_totalsuffix for the endpoint slice controller metric (#140447) - ✅ VS Code: center Modern UI panel title tabs in the 32px header (#331612)
- ✅ VS Code: Modern UI tabs reserve the close-button column so it doesn't overlay the filename (#330754)
- ✅ Appwrite: return ISO8601 session expire from
updateSession(#12891) - ✅ Playwright: pass client name on token-bypass connect (#41845)
- ✅ VS Code: fix Modern UI editor tab decoration color on full label (#325291)
- ✅ Apple Pkl: fix super method call inside a
letexpression (#1383)
✅ merged upstream · newest first
2025
- ✅ DigitalOcean Gradient:
wait_until_ready()for agent deployments (#56)
Projects
- antibody: every fix becomes a Semgrep rule your AI coding agent can't break, enforced on agent edits and in CI.
Python · Semgrep · static analysis · Claude Code - codereviewer_agent: multi-agent GitHub Action that posts only fact-checked, high-confidence inline PR comments.
Python · LangGraph · RAG · pgvector · GitHub Actions - a2a-incident-agents: incident diagnosis and remediation agents with approval gates, traces and evals.
Java 21 · Spring Boot · Spring AI · Postgres + pgvector - usage-billing-pipeline: exactly-once usage billing with Redis dedup, a transactional outbox and a DLQ.
Go · Kafka · Redis · PostgreSQL · Prometheus - CopilotPulse: local-first VS Code extension with dashboards for Copilot Chat sessions, tokens and per-model cost.
TypeScript · React · VS Code API - briefly: AI briefing agent whose durable workflow plans, drafts, revises and saves briefs, with persistent memory.
TypeScript · Cloudflare Workers AI · Durable Objects · Workflows
Marathons · weight training · too many terminal tabs





