Backend engineer building AI agents at scale: distributed systems, real-time streaming, message queues, zero-downtime migrations, and multi-region AWS.
I work on the AI Platform at Leena AI, where the systems I build carry every conversation between an enterprise employee and an AI agent, across 500+ enterprises: the streaming runtime, the agent auth, the channel integrations. When a message survives a pod restart or a dropped socket, that is my code doing its job quietly.
I build most of it with Claude at the center of my workflow. The review pipelines, test generators, and code search tools I wrote for myself now run for my whole team, and made our delivery cycles roughly 10x faster. I started out writing test automation at Cognizant, which is why I still write code assuming someone will try to break it. Along the way I earned an M.Tech in AI and Data Science from IIT Patna.
What happens between someone pressing send and an AI agent's reply. On the left, what they see; on the right, the runtime I build doing its job. Hit “drop the connection” to cut the network mid-reply and watch nothing get lost.
Done: the new greeting is live for every employee.
now · step 1 The admin's message reaches the runtime.
now · step 2 Token verified; this admin may change bot settings.
now · step 3 Conversation, available tools and instructions go into one prompt.
now · step 4 Words stream to the screen as they're generated.
now · step 5 It calls the platform API that updates the bot.
now · step 6 The rest streams back; a dropped connection resumes where it left off.
done Six steps, one reply. Try dropping the connection.
An AI operator agent inside the admin console. Streams responses over SSE, executes tools against platform APIs, and lets admins configure and debug bots conversationally. Exposed over A2A and MCP so external agents can drive the same workflows.
A tiered pipeline of Claude review agents with adversarial verification, adopted team-wide. Lower LLM token spend per review, and turnaround cut from hours to minutes.
Real-time SSE and WebSocket backbone: reconnect message replay, server-driven UI components, and agent token auth (JWT/ACL) on Node.js/TypeScript microservices across multi-region AWS.
Shipped WhatsApp v2, MS Teams provisioning, live-chat handoff with RFC 7662 token introspection, and SAML SSO, end to end.
A mandatory sign-off gate covering scalability, reliability, feature flags, and latency, backed by RabbitMQ service-level health checks and zero-downtime DB migration patterns.
Natural-language code search across the team's repositories, built on tree-sitter parsing, BM25, and vector embeddings.
A component design system and micro-frontend framework. Migrating legacy frontends to a modular monorepo cut build times, and Lighthouse-driven work improved load performance.
Three agents review every pull request the way a good team does: one decides what needs a look, specialists look, and a skeptic throws out anything that doesn't hold up.
Step 01
Decides what the change needs reviewed.
Step 02
Focused reviewers work only on what triage flagged.
+ MCP code search context from 9 repos, not just the diff
Step 03
Challenges each finding; weak ones are dropped.
Tap any skill and my agent shows where I've used it. Highlighted skills are what my production work runs on.
Own the real-time conversation runtime and agent auth behind AI agents for 500+ enterprises. Built the Agent Dashboard and the team's Claude Code + MCP toolchain, shipped 4 channel and auth integrations, led a 948-error type-safety migration, and run L3 support and releases for 3 product modules.
Migrated a legacy BackboneJS frontend to React, shipped new Chrome-extension features, and built end-to-end web test automation that cut manual QA effort.
M.Tech, Artificial Intelligence & Data Science Engineering. 8.80 CPI
Bachelor of Engineering, Computer Science. 7.80 CGPA
MongoDB certified: Building AI Agents, AI-Powered Search with Vector Search, RAG Apps
An agent trained on my profile. Ask it anything about me: it answers questions, hands over my resume, and can take you to any section of this site.
An agent trained on Avinish's profile. Ask about his work, experience, skills, or how to get in touch.
Open to Senior AI Engineer roles: AI agents, LLM systems and the backends behind them, and to conversations about LLM-powered developer tooling.