I am a software engineer who builds production AI infrastructure: agentic systems, LLM-powered backends, and the distributed plumbing that keeps them reliable. Most recently, I was a Software Development Engineer at Wittypen in Pune. There I built Zerply.ai from zero to production as its primary backend and GenAI engineer. Zerply is an AI visibility, SEO, and sentiment platform that tracks how brands show up across ChatGPT, Claude, Perplexity, and Gemini.
At Zerply, I built Venus, a multi-agent runtime on Cloudflare Workers and Durable Objects that powers the product's conversational and workflow agents. It has a core harness for session and state management, and a tool-proxy layer that securely forwards agent tool calls to backend services. I also designed a production-grade Model Context Protocol server with 22+ tools, custom authentication, Redis-backed token caching, and a hybrid OAuth 2.0 Dynamic Client Registration (RFC 7591) flow. It lets AI clients like Claude Desktop and Cursor discover and invoke backend capabilities on their own.
On the infrastructure side, I built distributed workflows on Temporal for prompt evaluation, recursive summarization, and platform monitoring, extending effective LLM context windows by 10Γ while cutting token costs by ~90%. I built an async semantic indexing pipeline that crawled and indexed 10K+ URLs into Qdrant for RAG, and an embedding pipeline on OpenAI Batch APIs that cut embedding costs in half. I also built a real-time chat engine over Server-Sent Events that routes across GPT, Claude, and Gemini, and took the FastAPI backend from ~1s to under 100ms per request.
On the side, I built a RAG Document Assistant, a source-grounded Q&A system that answers only from uploaded documents with file, page, and chunk citations. It handles versioned re-indexing so stale documents never leak into answers, drops citations the model invents, refuses when evidence is missing, and treats document text as untrusted to defend against prompt injection. On my eval set it hit Recall@5 of 1.00, answered 17/17 correctly, refused 4/4 unanswerable questions, and is covered by 71 automated tests.
I work mostly with Python, FastAPI, Temporal, LiteLLM, Langfuse, LangChain, Qdrant, Redis, AWS, and Next.js. My areas of interest include agent architecture, LLMOps, retrieval systems, and distributed systems, and I am currently going deeper into Kubernetes and Kafka. I hold a B.E. in Computer Engineering from Savitribai Phule Pune University (2024).
I am open to backend and AI engineering roles and can join immediately. Reach me on LinkedIn or at mahajanh006@gmail.com.



