tejas parthasarathi sudarshan

Tejas Parthasarathi Sudarshan

Founder, Winsen Labs · Neural DNA · Chennai

cv

Hello! I'm Tejas and I build things. Chartered accountant turned self-taught builder. I run an AI lab, publish research on how networks should grow, and hold strong opinions about music, movies, and markets, argued over happily.

01 Now · Winsen Labs

One lab, three products. Formerly FanDesk.

The idea hasn't changed since day one: when all of a company's context lives in one place, AI can actually work, because it reasons across everything instead of what's inside one app.

ProprietaryWinsen BridgeA workspace with a company brain underneath it. Mail, chat, tasks, docs, and calendar in one self-evolving model of the company, and AI employees doing real work on top of it, approval-first.winsen.ai →Open sourcePlatosThe open-source agent runtime. Build, ship, and operate AI agents on infrastructure you own. Model-agnostic, monitored by default, no lock-in.platos.dev →Open sourceRocketmanProject management for the age of AI coding agents. A hub that lives inside your repo as one offline file, plus a skill stack that walks an idea from spec to shipped.github →

Around the products: a few consulting engagements a year, and Winsen Academy, our school for the AI-native way of working. The lab lives at winsenlabs.com.

02 Also now · Fanpit

Helping brands build communities around them.

Fanpit is where it all started, and it's very much alive: an events and community platform, made in Madras. Brands and artists host on it, people find events that move them. Communities across Chennai, Mumbai, Hyderabad, and Kochi run their shows, open mics, and gigs through it, and every bit of that growth has been organic.

2,000+members
100+events hosted
100%organic growth

Find something that moves you at fanpit.live.

03 Research · Neural DNA

Networks should grow their topology the way brains do.

Default disconnected, with metabolic cost forcing selectivity. A tiny genome decides how the network grows, and the same idea travels from artificial networks to real biological ones. Three papers so far.

354genome parameters
35.4Mconnections wired
99,970:1compression
3/3benchmarks vs GPT-2
Paper 1A compact genome for growing network architectureGenomes of 226 to 374 parameters controlling up to 2.2M connections across MLPs, CNNs, transformers, and video models.Paper 2Scaling the genome to GPT-2354 parameters wire 35.4M connections, a 99,970:1 compression, beating GPT-2 on WikiText-103, Penn Treebank, and LAMBADA.Paper 3The same genome, on biologyPointed at the human protein interaction network, it rediscovers cancer pathways and two clinically validated tumor markers without supervision.

Start at the research hub , where you can watch the topology evolve in an interactive visualization. Papers on Zenodo, code on GitHub, genomes on Hugging Face.

04 The road here

Accounting first. Building second. Stubbornness throughout.

2019Audit at BSR & Co. IPOs, mergers, OEMs. Learned how to think about problems.
2022Chartered Accountant, ICAI. Final cleared in the first attempt.
2023KPMG, financial risk. Promoted ahead of track. Kept feeling the tools were the problem.
2024Left. Taught myself to build. No CS degree, no bootcamp.
2025Fanpit: community infrastructure for brands. Sony Music India artists, Chennai scenes.
2026FanDesk became Winsen Labs. The lab, the brain, the runtime, the research.

05 Sidequests

Small things, built for the joy of it.

SkillgraphMIDICodeLet's Find SanityKodartTapMeow

06 Words & taste

I write in public and keep score of what I consume.

BlogEssays and build notesWednesday WaffleThe weekly newsletterThings I readBooks, with marginsMoviesWatched and argued overfreewriteA timed editor I built