/about
I build real AI systems for real businesses.
I'm a senior engineer who has spent close to a decade designing and shipping production systems that move real numbers for real businesses. My core craft is distributed systems, cloud-native architecture on AWS, event-driven architecture, and domain-driven design. I model data for scale, not slides: clean bounded contexts, well-shaped events, and storage choices that hold up when traffic gets ugly.
On top of that foundation I build Agentic AI and AI-driven data solutions: agents that take action inside live business workflows, with policy, audit, and evaluation wired in from day one. The goal is never a clever demo. It is a measurable outcome: faster lead-to-engagement, higher recovery rates, lower cost to serve, decisions made in milliseconds instead of meetings.
Sadaf Labs is the response to a pattern I keep seeing: teams with strong product ideas and weak system foundations. We are a lab and studio. The lab ships open-source tools for trustworthy agents and durable data platforms. The studio takes on a small number of engagements where senior engineering moves the business, not the backlog.
How it started: PLANTICO
Long before AWS, agents, or RAG, there was a leaf. My final year university project, PLANTICO, used image processing to detect plant diseases from a single photo and suggest a remedy. The idea was simple: monitoring big farms by hand is slow, expensive, and catches problems too late. A camera and an algorithm could do it in seconds, at the earliest stage, for almost no cost.
PLANTICO was selected as one of the top 5 projects nationallyat the National Grassroots ICT Research Initiative (NGIRI) competition, run by the National ICT Research and Development Fund at Margala Hotel, Islamabad. The Minister of State for IT and Telecom was the chief guest. Country leaders from IBM Pakistan and the local startup scene walked the floor. As a final year student standing next to a poster about diseased leaves, that day rewired me. Software did not have to be abstract. It could be a tool you put in someone's hand and watch them use.
Everything since has been a longer version of that same project. Build a useful thing. Put it in front of people. Measure whether it actually helped. Repeat.
Principles
Ship, then iterate
Real users beat perfect specs. Production in weeks, not quarters.
Boring infra
Lambda, Postgres, S3. Not the hot framework du jour.
Trust by default
Every agent gets policy, audit, and replay before it sees real users.
Honesty over hype
If LLMs are wrong for the job, we'll tell you. Often they are.
Path so far
2016
Final year project PLANTICO. Image processing system that detects plant disease from a leaf photo and suggests a remedy. Top 5 nationally at the NGIRI competition by the National ICT Research and Development Fund, Islamabad.
2017
First engineering job. Building intelligent transit and fleet operations software for public transport operators. Real-time at scale, real users, real consequences.
2018
Shipped a one-click CMS and site builder for B2B companies. Learned how to design for non-technical users.
2019
Joined contact centre AI work. Real-time pairing and routing for tens of thousands of agents.
2021
Moved to AWS serverless. Led a fintech debt recovery platform from zero to production in Europe.
2023
Went deep on LLMs and agents. Shipped first RAG system at scale.
2026
Started Sadaf Labs. Lab tools (PolicyLint, Trace Replay, Eval Harness) in pilot with design partners.
What I am building next
Global Passport Chain. A trust layer for agent-to-agent commerce.
Picture this. Your agent wants to buy something on your behalf. A flight, a contract renewal, an API quota top-up, a bag of coffee. Today the merchant has no clean way to know if the agent on the other end is allowed to spend that money, on whose behalf, under which policy, with which guardrails. So the merchant blocks it, or worse, lets it through and finds out later.
Global Passport Chain is a public, verifiable credential chain for agents. The user issues a passport to their agent. The passport carries identity, scope, spend limits, allowed merchants, expiry, and a chain of delegations if the agent hands work to another agent. Every link is signed. Every link is auditable. The chain is public so anyone can verify it without calling a private API.
When the agent reaches checkout, the merchant or the payment processor (Stripe, Adyen, anyone) does a single verify. Is this passport valid. Is this purchase inside scope. Is the chain unbroken. If yes, the payment goes through. If no, it stops at the gate, with a reason the user can read.
Think of it as TLS for agentic commerce. Or a global notary that any agent, any merchant, any platform can read from without trusting each other first. Built on open standards so it does not belong to one company, including this one.
This is the long arc. Sadaf Labs is the studio. Global Passport Chain is the bet. Think of it as a global passport for an army of agents, one that any merchant can verify without trusting the agent or its owner first.