Aneeq Khatriteaching
Teaching agentic AI at national scale
I am Head of Faculty at the Governor Sindh Initiative for AI and Computing, where I lead around a hundred instructors and own academic quality across cohorts. I also train engineers at PIAIC and Saylani. Over five thousand engineers have learned agentic AI through these programmes.
The argument
Classroom AI and production AI are not the same subject, and most curricula teach the first while claiming to teach the second. A student who can get an agent to answer a question has not learned the job. The job starts at the question after that: how do you know it is right, what happens when it is not, and who is accountable when it acts.
So the curriculum is hands-on by design. Students build, break, test and ship real agentic workflows. Multi-agent architectures, tool calling, evaluation, the reliability practices I use in my own engineering work. They develop the practical judgment the industry actually asks for, rather than a portfolio of demos.
Running a faculty of this size is its own engineering problem. Quality assurance across cohorts means class synchronisation, content fidelity, syllabus adherence and progress tracking. And a standard that holds whether a student is in the first cohort or the tenth. That consistency is the part that does not scale by itself.
Classroom AI vs production AI
This is the comparison the curriculum is built around. The left column is what most courses teach and what most demos show. The right is what the same task looks like once someone is accountable for the output.
| concern | classroom | production |
|---|---|---|
| Correctness | It answered | It answered, and an eval says how often |
| Uncertainty | Confidence from the model | Confidence computed from data the model cannot write |
| Failure | Retry the prompt | Escalate, with the evidence attached |
| Tools | It called the tool | Tool-selection accuracy is measured and gated |
| Refusal | Treated as a bug | A first-class outcome, rewarded when correct |
| Change | Edit the prompt, ship | Prompt change fails CI until the eval passes |
| Accountability | The demo worked | A human signed the action that moved money |
Students build systems in the right-hand column. The projects on this site are the same argument at production scale.
Roles
Feb 2024 to Present
Head of Faculty & Lead Trainer · GIAIC
Governor Sindh Initiative for AI and Computing · Part-time · Karachi, Pakistan
- Lead a faculty of ~100 instructors across multiple cohorts, setting teaching standards and ensuring consistent delivery at national scale
- Trained over 5,000 engineers in agentic AI through Pakistan's national AI programmes
Aug 2025 to Present
Agentic AI Trainer · PIAIC
Presidential Initiative for AI and Computing · Part-time · Hybrid · Pakistan
- Teach agentic AI across national cohorts: agent architectures, tool and function calling, retrieval, and the evaluation practice that tells a student whether their agent actually works
- Ship reference implementations that are adopted across national programme projects, so the teaching material and the production patterns stay the same thing
Aug 2026 to Present
Lead AI Trainer · Saylani Mass IT Training (SMIT)
Part-time · Karachi, Pakistan
- Design and deliver hands-on agentic AI training covering multi-agent architectures, tool calling and agent evaluation
- Bridge classroom AI and production AI: students build, break, test and ship real agentic workflows rather than following demos
Speaking and training enquiries
I deliver hands-on agentic AI training covering multi-agent architectures, tool calling, retrieval, evaluation and the governance patterns that make agents deployable. If that is useful to your team, get in touch.