# Pavitar Rana > Backend & infrastructure engineer. I build the infrastructure AI agents run on. Early backend engineer at Mayson, where I own the FastAPI and Postgres backend and the Kubernetes layer that AI-generated apps deploy onto. On the side I'm building Fuse, a Firecracker sandbox runtime, from the hypervisor up. Location: India · UTC+5:30 Email: ranapavitar14@gmail.com Résumé: https://pavitar.dev//resume.pdf GitHub: https://github.com/pavitar-rana X: https://x.com/pvtrtwt LinkedIn: https://www.linkedin.com/in/pavitar-rana/ ## Experience ### Mayson, Early Backend Engineer (Dec 2025 – now) https://mayson.dev. Pre-seed platform that turns a prompt into a deployed full-stack app. I own the backend and the Kubernetes layer every generated app runs on. - **Job runtime.** RabbitMQ supervisor fanning one handler across many queues. Each job runs in an isolated process with memory caps and timeouts, so a bad job dies alone. [RabbitMQ · process isolation] - **Multi-tenant data plane.** Per-tenant Redis and S3 namespaces, presigned URLs, and Fernet-encrypted credentials injected into each app's Kubernetes Secrets. [Redis · S3 · Kubernetes] - **Realtime, end to end.** WebSockets through nginx ingress with KEDA autoscaling. One-click app remix streams progress over SSE while a fresh tenant database is provisioned. [WebSockets · SSE · KEDA] - **Agent integrations.** Gmail, Drive and Telegram via OAuth2 and webhooks, exposed as skills the agent reads and wires into apps on its own. Plus SMS OTP auth. [OAuth2 · webhooks · agent skills] ### Earlier - **Tenc**, Full Stack Developer · Backend, Infra & AI (2024 – 2025): Shipped applied-AI products for clients: RAG pipelines and LLM features on Node.js, FastAPI, Postgres and MongoDB. Owned Docker, CI/CD and deploys on GCP and Azure; mentored juniors. - **Axentia**, Full Stack AI Developer (2023 – 2024): Built an OCR-based AI math solver and a mental-health assistant on Flask and Node.js, with OpenAI and LangChain RAG. - **BITS Pilani**, B.Sc. Computer Science (2024 – 2027) ## Fuse: Self-hosted sandboxes on Firecracker microVMs https://github.com/pavitar-rana/fuse An E2B/Daytona-style runtime for running untrusted code, built from the hypervisor up. Every sandbox is a microVM with its own kernel, rootfs and network, created and driven over a REST API. - Firecracker driven directly over its Unix-socket API - Networking by hand: TAP per VM, bridge, NAT, iptables - Multi-tenant control plane on Postgres and Redis Next up: snapshot/restore, multi-host scheduling and quotas. Build log: https://x.com/pvtrtwt ## Projects - **Exness** (https://exness.pavitr.cloud/): Real-time CFD trading platform. Market data over WebSockets, Redis Streams feeding the order and pricing pipelines, high-frequency ticks aggregated in TimescaleDB for candles and history. [Node.js · Redis Streams · TimescaleDB · WebSockets] - **Chat Doc** (https://chat-doc-eta.vercel.app/): Split-view PDF reader with Gemini Q&A over embeddings-based retrieval. Azure Blob caching and virtualized rendering keep 300+ page documents responsive. [Next.js · Postgres · Gemini · Azure Blob] ## Tools - Languages: TypeScript, Python, Go, SQL, Bash - Backend: FastAPI, Node.js, PostgreSQL, Redis, RabbitMQ, TimescaleDB, MongoDB, WebSockets - Infra: Kubernetes, KEDA, Docker, Firecracker, KVM, Linux networking, AWS, GCP, Azure, GitHub Actions - AI: LangGraph, LangChain, RAG, agent tools and skills, OpenAI, Anthropic and Gemini APIs _Updated 2026-11-20_