About
Founder-led. Hands-on. No account managers, no junior bench, no handoffs. Every project gets all of us.
AI and backend · IIT Roorkee · Singapore
Seven years of production AI across medical imaging, visual search, connected vehicles and agentic systems. Builds the models, agents, retrieval and the infrastructure underneath. US patent holder and open-source MCP maintainer.
Product · builds FireOS · India
Turns ambiguous problems into shipped product, from PRDs to pixels to production. Builds FireOS, the compliance operating system for fire and life-safety agencies.
Founder · strategy and operations · India
Sets the direction and runs operations. Keeps the bar high on what we take on and what we promise.
How we work
The same four steps on every engagement, written down so you can hold us to them.
Step 01
You describe the week. We map the real workflow and the data, and say whether software is the answer.
Step 02
A written plan: what it may and may not do, who reviews what, the cost and the risks. You approve it.
Step 03
Tested on your real cases before it ships. You see progress every week and keep control of priorities.
Step 04
We measure the outcome against what we agreed and iterate. Running it stays cheap, and every action is logged and undoable.
Step 05
Maintenance, monitoring, documentation and handover.
Then it runs lean
Under the hood
Four things from real builds that show how we think about safety, reversibility and proof.
Model-written SQL is parsed into a syntax tree and must pass an allowlist (one SELECT, no writes, no personal-data columns) before it runs.
Read the case →
Every automatic data fix stores its before and after, so an admin can revert any single change months later.
Read the case →
Each agent has a recorded evaluation set that runs before any prompt or model change ships.
Read the case →
An open-source MCP server that connects AI assistants to 17 markets of regulatory data, used from Claude, ChatGPT and Cursor.
Read the case →
Built with
Ways to engage
All senior, all hands-on, in your time zone.
One AI system, start to finish
One well-scoped automation or assistant, shipped end to end with documentation your team inherits. Quoted up front.
We run your AI operations layer
An ongoing engagement where we own the automations and assistants end to end: building, operating and improving them against real traffic on a fixed cadence.
AI direction for your own team
You have people; you need direction. Tool and model selection, architecture, evaluation design and a frank read on what is worth automating.
Where we are
Ansuj is in Singapore; Vandan and Prashant are in India. We work in any time zone: our clients are in North America, Europe, Asia and Australia, and we keep their hours.
Credentials