About
A studio you can put a name to.
I build and ship the work myself. I started with machine learning and deep learning in 2019 — classifiers, NLP, Kaggle competitions, models deployed to cloud APIs — years before generative AI became a category, and I taught data science publicly while I was doing it. That foundation is why Rekvon does not stop at prompting: when a problem needs a trained model, a pipeline, or evaluation you can trust, we can actually build it.
You are not hiring a logo or a pool of contractors. You are hiring a named engineer in New Delhi who will tell you what your project actually needs — including when that is less than you asked for.
YouTube ↗
Aditya Pratap Singh
Founder & principal engineer · New Delhi, India
- Based in
- New Delhi, India
- Working hours
- US, EU, Gulf and APAC overlap
- Team size
- Small and senior — no juniors on your build
- In production
- ai.rekvon.in — our own AI speech platform
We were training models before prompting was a job.
Most studios selling AI today started when ChatGPT did. Ours started in 2019 with machine learning and deep learning — classifiers, NLP, Kaggle competitions, a model deployed behind a cloud API — and it has all been public the whole time. Open any of it.
- 2019
- Building machine learning since
- 81
- Public projects you can read
- 6
- Languages shipped in production
Classifiers, NLP and a deployed model API — before generative AI existed
Not forks — things built from scratch
Python · TypeScript · JavaScript · Go · Dart · Shell
Machine learning
Classification, regression and forecasting on your own data — trained, evaluated, and deployed behind an API rather than left in a notebook.
Deep learning
Neural networks for language, vision and audio where classical methods run out — including the training and evaluation loop, not just inference.
NLP & language systems
Text classification, similarity, extraction and retrieval — the groundwork underneath every retrieval-augmented product that actually works.
Generative AI & agents
LLM agents, voice and video systems, and model integration — built on the evaluation discipline the older work taught us.
ViralLabs
Source
Full-stack AI platform for YouTube creators — script generation, thumbnails with face cloning, and voiceovers on GPT-4 and Gemini. React 19 + FastAPI.
Open ↗smart-postgres-mcp
Source
An MCP server that lets an AI assistant query and reason over a Postgres database safely — the same integration pattern we build into client products.
Open ↗Fake News Classification
ML · 2020
A machine-learning classifier that flags likely-fake articles — one of several NLP and Kaggle builds from 2020, three years before generative AI became the default answer to everything.
Open ↗IriisApi
ML · 2020
An ML model deployed to a cloud API in 2020. Training a model is the easy half; serving one is where most machine-learning work quietly dies.
Open ↗Study Buddy
Live
A deployed TypeScript study app — AI-assisted learning, shipped and running.
Open ↗The Quran Guide
Live
A deployed reading and search interface built in TypeScript, live for anyone to use.
Open ↗
How we operate.
You talk to the person building it
There is no account manager between you and the code. The person who scopes your project is the person who writes it and the person who answers when something breaks.
Demos are not the finish line
Most AI work stops at something that looks good in a screen recording. Ours has to survive real inputs, real load, and the second week. Rekvon exists because we kept being handed the first kind and asked to fix it.
We run what we build
Our own product runs on our own infrastructure — we own its uptime, its latency budget and its bill. That is a different discipline from wiring up an API and handing over a repo.
Honest scope, honest price
Fixed scope for defined projects, monthly for ongoing work. If we think your idea is smaller than you think, we will say so on the first call rather than after the invoice.
Want to know if we're a fit?
Fifteen minutes, no deck. You describe what you are building and leave with a real scope and a real number — whether or not you hire us.
Free 15-minute teardown · reply within one business day