We started with a hunch: in AI, the hard part was never the model.
Everyone can demo AI. Almost no one can make it land inside a real organization.
That gap is the whole reason Canvas Labs exists.
We spent years building things for other people.
Himanshu ran Parallel HQ, one of India's largest design studios, shipping products used at national scale, from DigiLocker to CoWin.
Then ChatGPT arrived, and the ground under software moved. About a month later we started Canvas Labs to build on it. One early project became a second, then a third, and we understood what we actually were. A research and product lab.
We've shipped production AI since our very first week.
We have an itch: when a new model or update drops, we're the first to pull it apart and find out what it can really do.
Just before the reasoning models arrived, we were deep in fine-tuning. We built our own product to train and control models properly, because renting a black box was never going to be enough. Then o1 landed, and Claude collapsed the prompting game almost overnight. That was the signal. The ground had moved, and it wasn't going to stop.
Project after project, the same lesson surfaced. The model was never the moat.
We could swap one model for another in an afternoon. What held value was the organization's own context: its data, its decisions, its people, structured so an agent could actually use it.
On one engagement, fine-tuning both worked and failed at the same time. That contradiction made the answer obvious. We had been rebuilding the same primitive on every project. So we named it and built it properly. The Context Graph.
The future of AI is agentic.
As Satya Nadella, Microsoft's CEO, puts it, every organization will soon run on two kinds of capital, human and token, with people and agents working side by side. For that to hold, a company needs its own context graph: a private, structured model of everything it knows, running in its own cloud, even fully offline.
Models will keep changing, and access to the best of them is becoming political. Your context is the part you own. Engineer it well and you can swap the model underneath whenever you want.
You can't out-model your competitor for long. You can out-context them forever.
Palantir showed what a knowledge graph is worth to governments and the Fortune 50. We're making the same primitive available to every company.
We don't do this from a distance.
We forward-deploy, building the context graph inside your world, next to your people. A capability, not a vendor.
Distribution is part of the product.
Every serious software company is becoming a media company, so we build in the open and teach what we learn. You'll find it in our Research Notes, and on Himanshu's LinkedIn, Instagram, and YouTube. The research, the experiments, and the failures, shared as we go.
Canvas Labs is an applied AI lab, based in Bangalore, shipping production AI since 2023.
We're the lab behind Sketch, Canvas Connectors, and the Context Graph platform.
- Scaled Zomato's Hyperpure (as Business Head) 4× to $40M within 18 months.
- As Co-Founder, grew Parallel's digital product design topline 5× to $1.25M within 24 months.
- 18+ years working with data and analytics.
- Master's in Data Science, IIT Hyderabad.
- Led a 25-member engineering team to build Artemis, Travelopia's travel-planning software.
- 8+ years building high-performance applications (across orgs including Equal Experts).