Muhammad Waqas — AI and Software Architect. Ten years building systems for banking, healthcare, telecom, risk and compliance, where a confident wrong answer is a reportable incident.
A representative agent run: the agent plans six steps within a cost budget and timeout, searches the CRM, queries entitlements, hits a field mismatch on billing country and retries with a fallback, updates the CRM validated against a schema, passes a policy check, waits for human approval, and completes six of six steps in 11.4 seconds for $0.31 with the trace persisted.
Registered users on the healthcare platform I architected from scratch, now live across Libya.
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.
In their words
Written on LinkedIn by the people I reported to. Every name links to their profile — check any of it.
Waqas is a highly skilled technologist and possesses a breadth of knowledge that spans across multiple domains. His ability to work with different technologies and platforms was simply amazing, and he was always willing to take on new challenges that required learning new skills. He has a keen interest in staying updated with the latest advancements in the software industry.
In addition to his technical expertise, Waqas was an exceptional technical mentor to our junior engineers…
Waqas worked with me for 2.5 yrs. approx. He has good work ethics, he always managed his goals well, has a well-planned career growth path. In the past 2 years he improved a lot on his skill set, task deliveries, work quality.
He is a keen learner and has gone over steep learning curves at the same time delivering good quality work adding great values in the product. He is one of the very good team players and always offered his help and guidance to his colleagues when required.
I know Waqas as a hard working and very serious team player. He is result oriented and responsible employee and he is always ready to put all his energy and time to get the job done. I recommend him and wish him success in his future endeavors.
Quoted as written. Read them in full on LinkedIn (opens in a new tab).
Have a system like one of these in mind?
Stack
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.
Free
Two pages. The questions I ask before quoting any automation work — which processes are actually worth automating, what breaks once real data hits them, and the four checks that predict whether an agent survives contact with production.
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.
So you can size the conversation before writing: the scoping pass is a fixed $145 and takes about three days. A single production agent or retrieval pipeline — built, evaluated, deployed and handed over — typically starts around $850. Ongoing ownership of a live system runs from $1,400 a month. Larger or multi-system work is quoted from the scoping pass. If you would rather talk before writing anything, book a call.
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.
You do — all code, prompts and documentation transfer to you on final payment. I sign your NDA and MSA, and you can contract directly or through Fiverr or Upwork escrow, whichever your procurement prefers.
I lead AI product development at DTCForce, and I take on a small number of independent engagements alongside it. That shapes what I accept: scoped builds with clear boundaries rather than open-ended staffing, and I turn down anything that overlaps my employer's market. Client work runs entirely on my own time and tooling — no employer code, infrastructure or IP touches it, in either direction.
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, and you’ll get a straight read on whether it’s worth building. Prefer to talk it through? Book a time directly.