Skip to content

I take the data and AI work off your team's plate.

I am Farhan Ahmed Khan. I spent six years at a US-based data platform company, building and shipping products for enterprise clients around the world. Now I build AI systems for businesses directly, through my own company, Quartate.

  • Founderof Quartate, since August 2026
  • Six yearsat a US-based data platform company, building and shipping products
  • Last rolePrincipal Product Solutions Engineer, 2025 to 2026
  • 15 peoplein my team at its largest, across product and pre-sales
  • Four continentsNorth America, Europe, the Middle East and Australia
  • Employee of the Year2023, at the same company
  • LUMSBS Electrical Engineering
See what I have builtWhat I take on
Farhan Ahmed Khan, photographed against a pale grey wall

Data and AI bootcamps for more than 1,500 people, at top universities across Pakistan, among them IBA, NED, LUMS, GIKI and NUST. I also teach Research and Study Skills part-time at Lincoln.

  • Speaking at a lectern with a microphone, gesturing mid-sentence, a data modelling tool projected behind
  • A full university auditorium of attendees with laptops open, seen from the front of the room
  • Teaching a bootcamp to a room arranged in a horseshoe, arm out mid-explanation
  • Presenting beside a projected data grid, hand raised mid-point
  • A long training room seen from the back, every seat taken and every laptop open
  • Mid-sentence in front of a seated workshop, hands shaping the point
  • The team on the steps of a university building before a bootcamp
  • Holding the Employee of the Year certificate and trophy at a company town hall

AI should only decide what someone can check.

Put AI somewhere a mistake would go unnoticed and it will fail without telling you. Most of my work is deciding where that line sits. I published how I draw it as a free, open-source skill, judgment-or-code.

The AI decides

Messy input, and a person can check the answer.

A person approves

Money moves, or it cannot be taken back.

Code does it, not AI

A mistake here would be costly and easy to miss.

Why it sits there

Messy input, and every guess it makes can be checked against the data before anyone builds on it.

Pick one, or leave it running.

What I take on

Quotes and orders from email

When a quote request or an order arrives as an email or a PDF, AI reads it and matches it to your products, and you approve it before it goes out.

Purchases onto the right job

Receipts and supplier invoices are matched to the job they belong to, so the costs you can bill reach the customer's invoice.

Work that never got billed

Finished jobs are checked against your contracts and invoices, so you find the work nobody invoiced.

AI doing the repetitive part

I build agents that take on the work nobody wants to do twice, and a person approves anything that matters.

Recent client work

Quoting and approval for Setlexor

Working with Peace of Growth, I built the quoting and approval system for Setlexor: proposals that took about three days now go out in minutes, after one approval from their commercial director.

Course sourcing for Lincoln International Foundation

I built a system that helps their team use AI to gather educational sources and bring them into their course material, faster.

A company knowledge base for Lincoln International Foundation

Every meeting, recorded in Urdu, English or both, is kept and indexed in one knowledge base. Admins ask questions across every meeting, and each answer names the meetings it came from.

Things I have built

Talking to a data platform

Enterprise product

I led the work that let customers describe the data setup they wanted in plain English and have the software build it.

The call I made. I led the design and set its limits, so anything it is not allowed to do gets a clear no instead of a confident wrong answer. Engineers built it.

Months to days
for a customer to get up and running
Hours to seconds
to build a working data pipeline
MCPAgent skillsETLData modelling

How I work

Write it down until it is unambiguous

I get the agent to interrogate me about a problem until nothing is left vague. Vague requests are where most of it goes wrong.

More than one AI on the hard jobs

I build with Claude Code and bring in Codex and Cursor when a job is hard. Deciding how they hand work between them is its own skill.

I test everything it produces

I do not hand-write production code. I build with Claude Code, then test the result against the job it has to do.

Nothing marks its own homework

Whatever matters gets checked by a different AI than the one that wrote it, and plain facts get checked by code.

If your business runs on work someone does by hand, I would like to see it.

Work with me at Quartate

or write to me, farhan1188@gmail.com