A small studio, on purpose
Smartgenhub is a handful of engineers who build machine learning systems for other teams. We stay small because it keeps us honest: no account managers, no upsell targets, just the people doing the work.
Why we started this
We kept watching the same story play out. A company gets excited about AI, hires a big consultancy, and six months later has a beautiful slide deck and a model that never leaves the demo environment. Meanwhile, the actual problem — the one a focused team could have solved in two months — is still sitting there.
So we decided to do the opposite. Small scope, real data, working software. That's the whole pitch.
We won't claim to have hundreds of clients or a wall of awards. What we can offer is careful engineering, plain communication, and a refusal to promise things machine learning can't deliver.
We say no a lot
If your problem doesn't need ML, we'll recommend the simpler thing — even when it means a smaller project for us.
We show our work
Every model ships with an evaluation report: where it works, where it fails, and how often. No cherry-picked demos.
We build for handover
Your team should be able to run what we build without us. Documentation and knowledge transfer are part of every project.
Meet the people behind Smartgenhub
A small leadership team that stays close to the work. Everyone here writes code, reviews models, or talks to clients directly — no layers in between.
Rahul Sharma
CEO & Co-founderSets the direction and keeps every project honest about scope and outcomes. Spent a decade shipping data products before starting Smartgenhub, and still joins the first call with every client.
Priya Mehta
CFOOwns budgets, pricing and compute costs. Makes sure GPU hours are spent where benchmarks justify them, and that clients always know what a project will cost before it starts.
Arjun Singh
CTO & Co-founderLeads architecture and engineering standards. Reviews every system we hand over — from data pipelines to deployment — so client teams can run what we build without us.
Sneha Kulkarni
AI EngineerBuilds and evaluates the models: fine-tuning, computer vision, NLP and the evaluation reports that ship with every project. Believes the boring baseline should always be tested first.
How a project unfolds
Five stages, each with a clear deliverable. You can stop after any of them. Our own product is currently at Stage 03 — Prototype.
Data review
You share a sample of your data under NDA. We spend a few days with it and come back with an honest read: is there enough signal here for ML to help? Sometimes the answer is no, and the engagement ends there at no cost to you.
Scoping
Together we define one narrow, measurable goal. Not "add AI to the product" — something like "extract these five fields from invoices with under 3% error." Narrow goals get shipped; vague ones get presentations.
Prototype
Three to eight weeks of building. You get a demo at the end of each week — sometimes rough, always real. We'd rather show you an ugly working thing than a polished mockup. During prototyping, we use NVIDIA GPUs for model fine-tuning, computer vision, NLP, and demanding inference when benchmarks justify them; lighter pipelines stay on CPUs to keep costs controlled. If our NVIDIA Inception application is approved, eligible cloud credits would support these benchmarked prototype workloads.
See how we use NVIDIA hardware →Production
If the prototype earns it, we harden it: proper APIs, tests, monitoring, cost controls, and a deployment that fits your existing infrastructure rather than replacing it.
Handover
Documentation, evaluation reports, and working sessions with your engineers. We stay available for maintenance if you want us, but the system is yours and runs without us.