Intelligenceisn'twhatyouknow.It'swhetheryouknowthatyouknow.
Building AI that measures how humans think — not just whether they're right.
Karachi, Pakistan
It started with a question no textbook thought to ask.
Karachi, 2024
Why do students fail even when they study? Why does knowing the answer ≠ understanding it?
# The gap between these two values is what I research.
15 research phases later, HCMS was born. A DOI-backed preprint. A formal framework. At 16.
I don't build AI to follow a roadmap.
I build it because the problem is real and someone has to go first.
The Wrong Question
Data ≠ Understanding
A model trained on 10 million examples doesn't understand any of them.
Prediction ≠ Understanding
Getting the right answer doesn't mean knowing why it's right.
Accuracy ≠ Understanding
97% accuracy can coexist with 0% cognitive stability.
Understanding = Confidence Calibration + Reasoning Consistency + Cognitive Stability
This is what HCMS measures.
Read the Research →Human Cognition
Measurement System
"Beyond Correctness: Measuring Cognitive Stability
and Confidence Calibration in Human Understanding"
Shahid, M.R. (2026). Zenodo.
DOI: 10.5281/zenodo.18269740
Every test you've ever taken assumed correctness equals understanding. HCMS proves it doesn't. Across 15 structured research phases, HCMS models the gap between getting something right and truly knowing it — measuring confidence calibration, reasoning consistency, and cognitive stability under pressure. This is what assessment looks like when the question matters more than the answer.
At 16, that framework became a DOI-backed preprint. The research isn't finished — it's just begun.
Confidence Calibration
Measures the alignment between a student's confidence and their accuracy. Misalignment indicates overconfidence or underconfidence, both signs of unstable understanding.
Calibration gap visualization
Research Contributions
Introduces cognitive stability as a measurable dimension beyond correctness
Demonstrates confidence–accuracy misalignment predicts reasoning degradation
Provides diagnostic framework vs predictive scoring model
Interpretable, reproducible signals for education & cognitive research
Includes sub-systems: Cognitive Robustness Benchmark, Learning Analytics Engine, Confidence Calibration Module
Three Laws of Understanding
Law I "Understanding requires more than correctness."
Law II "Confidence without calibration is noise."
Law III "Intelligence that cannot explain itself is incomplete."
This framework is open-source and citable.
Shahid, M.R. (2026). Beyond Correctness: Measuring Cognitive Stability and Confidence Calibration in Human Understanding.
Zenodo. DOI: 10.5281/zenodo.18269740
Featured Projects
23 repositories. 7 deployed systems. 1 published preprint. Here are the ones worth your attention.
Human Cognition Measurement System (HCMS)
A 15-phase, research-grade cognitive assessment framework that goes beyond correctness. Models confidence calibration, reasoning consistency, and cognitive stability. DOI-backed preprint published on Zenodo.
DOI-Backed Preprint
Proof, not promises.
Every number below is a shipped output, a published result, or a real system.
Years old. Building what most wait decades to attempt.
In HCMS. Not iterations. Structured phases.
On GitHub. Every one shipped.
DOI-backed. Zenodo. At 16.
Fake News Detector. Real-world data.
Real inference. Real users.
What I build with.
I don't list skills I've read about. Every tool here has a GitHub commit or a published paper behind it.
Hover nodes to explore connections
Thinking Out Loud
Research notes, half-formed ideas, and questions I can't stop asking. On Substack.
Muhammad Rayan Shahid on Substack
Independent researcher working on human-centered AI and cognitive measurement. Interested in how understanding, confidence, and stability can be formally measured — not assumed.
The Manifesto
Most people spend years preparing to do research.
I started doing it.
At 16, I published a DOI-backed cognitive science preprint — HCMS, the Human Cognition Measurement System. Not because a professor told me to. Because I realized that every exam I'd taken was measuring the wrong thing. Correctness is easy to fake.Deep understanding isn't.
I work at the intersection of machine learning, cognitive science, and human-centered AI. My research asks: can we formally measure how a person understands something — not just whether they answered correctly? HCMS is the first answer to that question.
My thesis is simple: intelligence is a stability, not a score. A calibration. A consistency under pressure.
I'm not building AI to get a job.
I'm building things that don't exist yet. That's the only reason worth having.
229 contributions in 2025 · Joined GitHub Jun 2025 · 23 public repos
Let's do something
that matters.
Researchers, universities, collaborators, people who read HCMS and had thoughts — reach out. I read everything.
Response time: usually within 24 hours.
Preferred topics: research collaboration, academic opportunities, AI systems.