Anindya Chowdhury
Strongly consistent code. Eventually consistent life.
I build software, ship it, and keep it running. At Videoverse I own the partner integration layer for Magnifi: the APIs, metadata, and event pipelines that connect us to some of the biggest broadcasters in Europe and the US. Away from work I build my own products and do the whole thing myself, down to the payments and the servers.
Lately I've been itching to build things that aren't software too. Mushroom farming and a food business, both early days, but I mean them. Outside that it's anime, manhwa, history rabbit holes, and embarrassingly long Valorant sessions.
A multi-agent app that reads Indian news so you can ask it questions. Ask how the papers are covering a story and you get a real answer, with every claim traced back to the article it came from. If nothing in the corpus fits, it says so instead of making something up.
It does real retrieval, hybrid search with reranking on top of vector lookup, over a corpus that's harvested around the clock. A Critic agent throws out anything it can't back up with a source. It's live, with accounts and payments, running on GCP.
Projects
All projectsWanted to see how far multi-agent LangGraph would go on Cloudflare's full stack in a weekend. Upload a photo of your food — three agents (Vision, Roast, Battle) argue over it and generate a shareable card.
Cloudflare Workers · Cloudflare Pages · Cloudflare D1 · Cloudflare R2 · LangGraph · Gemini
Wanted a reliable way to aggregate Indian news without hardcoding every source. Pulls from YAML-configured providers, enriches with goquery, deduplicates via BoltDB, and publishes to SQS, SNS, or Pub/Sub depending on where downstream needs it.
Go · AWS SQS · AWS SNS · GCP Pub/Sub
Writing
All postsNotes from building Samvad: a four-agent pipeline that answers questions about Indian news coverage and shows its sources. What worked, what surprised me, and what I'd change.
A short reflection on Agency, Consistency, and Ego — a simple framework that’s helped me shape goals and momentum, and might do the same for you.
Experience
Full history
Spent most of my time here building and owning the Partner API Service, a metadata orchestration layer that normalises AI-generated data across partner integrations. At peak it handles 150K+ events a day and directly changed how fast new partners could go live.
Other things I built: StreamForge (HLS playlist assembly across heterogeneous MAM storage providers), a match-schedule ingestion pipeline covering 100+ tournaments and 15K+ players from scratch, a live dual-write migration syncing MongoDB to PostgreSQL via SNS at sub-100ms propagation, and external publisher integrations to Dropbox, OneDrive, Slack and others for enterprise media workflows.
I also own P0/P1 incidents on partner-facing systems, introduced RCA practices that reduced recurrence, and led the monitoring migration from Datadog to Last9 + PagerDuty.
NestJS, Go, TypeScript, PostgreSQL, MongoDB, Redis, AWS.
Built a webhook-based calendar sync for Google and Outlook that pushed AI-generated post-call summaries straight into Google Docs, ClickUp, Notion, and HubSpot. Drove a ~25% engagement lift, which was a solid outcome for what started as a focused integration project.
Also shipped a Telegram voice chatbot running Pygmalion-7B with ElevenLabs for voice synthesis, served via AWS Lambda behind an ELB. That one was genuinely fun to build.
Python (Django), MariaDB, Redis, Supabase, AWS Lambda/EBS/SQS, React, Slate.js.
Reading
Everything I'm readingReading this to understand what talking to potential customers actually looks like. Something that barely comes up when you're on the engineering side. The way it reframes what counts as a useful conversation has already changed how I ask questions.
Solid foundation for thinking about software architecture. Read it even if you don't like his tweets. Especially then.
Always happy to chat about systems, side projects, or good food.
anindya.chowdhury@proton.me