Muhammad Waqas — AI Agent Engineer and Technical Lead. Ten years building systems for banking, healthcare, telecom, risk and compliance, where a confident wrong answer is a reportable incident.
Track record
Healthcare platform architected from scratch, now live across Libya and the Middle East.
Faster at a global market-research firm after rebuilding the development workflow.
Shipped end to end — agentic automation, risk, compliance, health, consumer insights, banking CRM.
In regulated industries where auditability and correctness are the requirement, not a preference.
What I build
Most AI projects stall the moment they meet real data, real load and real consequences. That transition is what I do.
LangGraph agents and MCP servers that carry out real operations end to end — with retries, guardrails, cost ceilings and a trace you can audit when something goes wrong. I designed an agentic platform that automates an entire Salesforce implementation from scratch.
Retrieval pipelines where quality is measured, not assumed — so the answer is still right on the tenth question, not just the one in the demo.
Where AI belongs in your system — and where it plainly does not. A decade of enterprise architecture behind the answer.
Snowflake, MindsDB and Airflow pipelines that feed models something trustworthy instead of whatever landed in the warehouse.
Banking, healthcare and compliance systems: access control, audit trails, and the paperwork that makes a security review survivable.
How an agent is actually built
This is the shape of every production agent I ship. The interesting parts are not the model call — they are the guardrails, the state, and what happens on the paths where things fail.
Selected systems
An autonomous agentic platform that performs a full Salesforce implementation — operations that took consultants weeks, executed and verified in minutes.
Enriches enterprise KPIs with macroeconomic and market data, then links them to objectives, risks and risk-appetite thresholds so emerging problems surface before they land.
Digital health infrastructure built with governments and health systems to widen access in emerging markets — health, education and advocacy in one localized platform.
Consumer insights across panels, surveys and custom research — the “Full View” of who buys what, how and why, used to anticipate demand shifts.
A cloud platform that automates compliance workflows and builds AI-augmented relationships between risk, regulation and organisational activity, surfaced through Power BI and Tableau.
A digital engagement platform driving customer journeys across every touchpoint, with workflow automation and lead generation for banks.
Stack
Credentials
Working together
You describe the problem. I tell you whether AI is the right tool, what it would take, and when the answer is “don’t” — before either of us commits.
Working code with the edge cases handled, the costs measured, and behaviour you can verify. Progress you can see weekly, not a reveal at the end.
Documented, tested, deployed, and explained — including the reasoning, so the next engineer isn’t guessing at why it works.
Questions
Most start with a short paid scoping pass: I look at your system and data, then come back with what is buildable, what it costs, and the risks. From there it is usually a fixed-scope build with weekly checkpoints, or an ongoing retainer if the surface keeps growing.
Almost certainly. I have shipped production systems in Java/Spring, Node, Python and PHP, across AWS, GCP and Azure, on Postgres, MySQL, MongoDB, Oracle and SQL Server. Agents get added to the system you already run rather than replacing it.
The way regulated industries require: least-privilege access, no client data in prompts unless contractually cleared, audit trails on every agent action, and a validation gate before anything with a side effect executes. A decade in banking, healthcare and compliance shaped that default.
Then I say so in the scoping pass, and you have saved a budget. Plenty of problems that look like AI problems are really data-quality or workflow problems — and I would rather tell you that than bill you for an agent that papers over it.
Pakistan (UTC+5), working regularly with teams in the US, EU, New Zealand and the Middle East. Meaningful overlap with EU mornings and US mornings; async by default, with responses usually within a day.
Yes — that is most of the last five years. I currently lead a product development department reporting to a CEO, and before that led engineering teams through weekly and quarterly release cycles for regulated customers.
Start here
One message with the actual problem is worth more than three calls about capabilities. Describe what is slow, manual, or unreliable — you’ll get a straight read on whether it’s worth building.