Plana
Plan a dinner with friends, order together, split the bill — one app.
- Flutter for iOS, Android and web, one codebase
- Shipped to the App Store and Google Play
- Bill split with GST, service charge and vouchers
01 / 12
Open the projectSrinath H · New Delhi · open to relocation
Full-stack AI engineer. I build products end to end — and train, evaluate and ship the small models inside them.
Interlude · space + time
Here are 32 frames of a SkyControl reel. Seen face on, it’s just a picture.
Turn it. Two dimensions are space. One is time. Every frame is a slice of the block.
Scrub the playhead and the past peels away. Any slice you stop on is a moment.
Cut it lengthways and you see change itself. That’s how I read a product — and a career: as one block, sliced.
Chapter I · Jan 2023 – May 2026
At PLXN World I was the founding engineer and the only developer. I shipped a social dining app, a tutoring marketplace, a distribution backend, an AI tutor and a live meeting companion — design to production, then on call.
Plan a dinner with friends, order together, split the bill — one app.
A tutoring marketplace for IB, IGCSE and A-Level — bookings, credits and payouts in one place.
A multi-tenant B2B distribution backend. The first version of a chain I run today.
A syllabus-aware AI tutor, grounded in the student’s textbooks and their own past lessons.
A meeting companion that transcribes English, Hindi and Hinglish live — and flags claims that don’t add up.
Chapter II · Oct 2026 – present
BlueRickshaw imports food and drink into India. I built its new platform from May 2026 and have led engineering since October: from a supplier abroad to a kirana shelf.
The import chain
Chapter III · Aeon Labs · May 2026 →
Where I think software is going, built in my own time and released in the open: models small enough to run in a browser tab, and tools that live inside the chat.
Voice air-traffic control at SFO, running on small models I trained. No LLM, no GPU at run time.
model out → UAL214 C 5000
An MCP server for honest options backtesting. Describe a strategy in plain words inside ChatGPT, Claude or Gemini; it’s tested on real one-minute prices.
“The service never calls a strategy good.”
What I do
01
Train from scratch, fine-tune (LoRA or full), RL with verifiable rewards (GRPO), then run them in the browser with ONNX on WebGPU or WASM.
02
RAG, agents, MCP servers, voice pipelines and document intake — wired to real users.
03
Next.js and React, FastAPI, Node, Flutter. Postgres, Redis, ClickHouse, Firestore.
04
Docker, Nginx, CI/CD with OIDC publishing, Prometheus, Grafana and Loki, security reviews.
Base
New Delhi, open to relocation.
BCA, Vellore Institute of Technology, 2020–2023.
Stack: TypeScript, Python, Dart, PyTorch, Hugging Face TRL/PEFT, ONNX Runtime Web, vLLM, LangChain, MCP, Gemini, Claude, Whisper, Next.js, React, FastAPI, Node, Flutter, three.js, PostgreSQL with pgvector, ClickHouse, Redis, Firestore, Docker, Nginx, GitHub Actions, Playwright.
Index







Local time, New Delhi
--:--:-- IST
Currently building
The BlueRickshaw platform, as Head of Engineering
Last shipped
SkyControl v1 · 14 Nov 2026
Stratify 1.0.3 · 13 Nov 2026