About

I'm Arbaz, a software engineer working remotely from India. For nine years I've built the unglamorous parts of consumer products, the streaming servers, the message queues, the payment flows, and the Kubernetes clusters that hold it all up when traffic arrives.

Since 2024 most of my work has been around AI models. GPU inference infrastructure, fine-tuning, and pipelines that run without people. I test what I learn by shipping. As an indie dev this year I put out a language model, an autonomous media channel, and apps.

Work

Turnip

Principal Software Engineer, 2020 to present

One startup, four pivots: mobile-game streaming, Clubhouse-style live audio, a creator social network, and now Zaps, an AI design suite.

  • Built and scaled the backend to 25M users, alone for the first two years. A 21-service Kubernetes fleet behind Kong, Kafka everywhere.
  • Live media at real scale. 50K streams ingested a day at sub-5s latency, and a live-audio platform with 2M monthly users and 70M call minutes.
  • Cut GPU costs 80% by moving AI generation to distributed GPUs with 2s image delivery, and infra costs 70% with autoscaling and serverless.
  • Architected the social platform. A dynamic news feed with second-degree discovery on Neo4j, real-time content indexing on Elasticsearch, and an event-driven core on Kafka.
  • Built RAG-based semantic search across the template library, embedding every design, carousel, and AI template for vector retrieval.
  • Built the AI enhancement pipelines for the Zaps era. ComfyUI face swap and retouch, orchestrated with n8n on Kubernetes, which took a large share of async operations off people's hands.
  • Built an agent platform where ten product people shipped 17 internal tools without a single engineer commit.
  • Migrated core services to Go behind a byte-parity strangler rewrite held at 100% test coverage.

Paytm Insider

Lead Backend Engineer, 2017 to 2020

  • Led the team behind movie ticketing that held 500+ requests a second on sale nights without a glitch.
  • Designed the logistics architecture. Delivery opt-in, order tracking, and the internal tools for managing shipments.
  • Built the full e-commerce flow. Cart, payments, inventory, and taxation.
  • Ran it all as AWS microservices on Kubernetes, with serverless functions on Lambda and Kinesis carrying the on-demand work.
  • Led a team of 7 and interviewed 100+ engineers for backend and DevOps roles.

Ithaka

Intern, 2016 to 2017

A travel-planning startup, and my first real-time systems work: built its omni-channel chat server bridging the app, Facebook Messenger, and support agents over Socket.IO and Firebase, with conversation handoff and Elasticsearch-backed search.

Provakil

Intern, 2016

Legal-tech. Wrote scrapers and fetchers that pulled daily cause lists from Indian courts, including the Delhi High Court and the national consumer courts, into structured case data.

Now

Right now I build as an indie dev. Not just the tech but the whole product lifecycle, from idea to distribution and growth. Most of it runs on open-weights models, squeezed to get high-quality automation at the lowest possible cost. Most days go into building automation pipelines, pushing current LLMs and diffusion models to find the cheapest way to run something autonomously without losing quality. RAG is one part of that work, with pgvector, local embeddings, and multimodal ingestion. Nothing here is a demo. Every project has been validated by real community response and real numbers. I work from India and overlap comfortably with US mornings and European afternoons.