Optivus · Company

Two friends.
One belief about how
AI actually ships.

Optivus is a small, senior team that helps companies turn AI from a slide deck into a system in production. We do the strategy and the engineering, the same people, the same room, the same week.

Founded 2025·New Delhi, India·10+ deployments
gt '22 · optivus '25
Advik Jain
advik · co-founder
Udayan Ambwani
udayan · co-founder
met first sem · 2018
roommates 4 yrs
→ optivus 2025
2 founders·1 dorm room·4 yrs of trust
Our story

How a college
friendship became
a company.

Advik and Udayan met in their first semester at Georgia Tech. They became friends almost immediately, roommates the next year, and then for four years after that.

After college, they spent the next few years at different companies, watching the same thing happen over and over: AI projects with brilliant strategy decks that never reached production. Pilots that worked in demos but quietly died in pre-prod. Roadmaps that took six months to write and never got built.

They kept comparing notes. The pattern was the same every time: the people writing the strategy weren't the ones building the system, and the people building the system weren't in the room when the strategy was written. So the strategy was wishful, the system was generic, and nothing shipped.

They started Optivus to fix exactly that. Same team for strategy and delivery. Senior engineers, in the room, on the work. A 2–4 week pilot that lands in production, not a 12-week PowerPoint that lands in a drawer.

We are still small. We plan to stay that way.

How we operate

Six commitments we make
before every engagement.

Not a values poster. These are the specific behaviors we hold ourselves to. The ones a client could hold us to.

01

We scope tight, on purpose.

Every engagement is one workflow, 2–4 weeks, with a written commitment to scope and timeline. If we can't scope it that way, we'll tell you why.

02

We commit to demos, weekly.

A working artifact every Friday, even on week one. Progress should be obvious without a status meeting.

03

We bring the eval before the model.

Before writing a prompt or fine-tuning anything, we write the eval suite that tells us when we're done. No vibes-based shipping.

04

We hand off, not throw over.

Engagements end with your team operating the system. Runbooks, training, on-call rotation: included, not extra.

05

We say "don't use AI" when we mean it.

A SQL view, a Zapier flow, or a better form is sometimes the right answer. We'd rather scope you out than scope you wrong.

06

We ship under our own roof first.

Veritas, Janus, FlowFin: real systems we operate at scale. The patterns we use for you are the ones we live with.

The team

Two founders.
Best friends since
freshman year.

One does strategy and the engineer-side of business. The other architects everything we ship. We trade roles depending on the engagement, but every call comes from one of us, in the room, on the work.

Advik Jain, Co-founder & CEO of Optivus Technologies
advik.jpg
Advik Jain
Co-founder & CEO
Strategy, partnerships, and the engineer-side of business. Previously on EY's Intelligent Automation team, building enterprise automation systems. Georgia Tech ME '22.
Udayan Ambwani, Co-founder & CTO of Optivus Technologies
udayan.jpg
Udayan Ambwani
Co-founder & CTO
Technical architect for everything we ship. Deep in the model, evals, and infra. Has spent more time looking at production traces than is healthy. Georgia Tech CS '22.
We're hiring soon
Looking for a third: a senior AI engineer who's shipped models to production and wants to do strategy and the work.
Get in touch
Where we work
Based in New Delhi.
Working with teams everywhere.
HQ
New Delhi, India
Founded
2025
What we're chewing on

What's on the desk this month.

We're not pretending to be neutral about this stuff. Here's what we're actually reading, building, and arguing about right now.

Reading
The Bitter Lesson, twelve years on
Rich Sutton (revisited)
Essay
Watching
How Anthropic builds with Claude
Engineering blog
Practice
Building
Eval suites that survive prod traffic
Internal RFC
In progress
Wrestling
When does fine-tuning actually beat prompting?
Open question
Honestly, often it doesn't