Founder & CEO of Nodes · Independent AI researcher
I am the founder and CEO of Nodes, the intelligence layer and system of action for enterprise decisions. My independent research spans multi-agent systems, the mathematical foundations of AI, persistent memory and context, and future compute architectures.
What I am building
Nodes runs inside a customer's cloud and reads the systems its teams already use. It surfaces a high-value open decision, drafts the workflow across those systems, and waits for a person to approve, edit, or reject the action.
Every pass creates a Decision Trace that preserves the evidence, source, model rationale, human reasoning, action, and outcome. Those traces accumulate into institutional context the customer controls.
Hiring is the first deeply validated wedge. The same architecture extends to workforce, operations, and revenue decisions.
900K+
candidates scored
10,765
hires analyzed in the production study
127 → 38 days
time to hire
47 days faster
to the first production milestone
Public figures as of July 2026. Sources: Nodes case study and arXiv:2604.19819.
In customer-VPC and on-prem deployments, core training, inference, scoring, and traces run inside the customer's environment on weights the customer owns. Deployment and security detail lives on nodes.inc/architecture.
Research that ships
The loop matters more than the label. I start with a question, formalize it, build or simulate the system, test it against reality when possible, and publish what the evidence supports.
Some of the work is production validated. Some is applied R&D. Some is intentionally exploratory. I label those categories separately because an idea should not borrow credibility from evidence it has not earned yet.
How models reason across disconnected systems, coordinate specialized work, preserve provenance, and turn recommendations into human-approved actions.
How intelligent systems represent state, preserve useful context, update beliefs, reason under incomplete information, and accumulate knowledge without losing the evidence behind it.
How to measure the value of acting, the cost of waiting, and the evidence required before an automated system can safely propose a consequential action.
Long-horizon work on architectures that may move beyond the limits of conventional silicon, including photonic, energy-based, and brain-inspired approaches.
Featured research
A study of 10,765 hires across applicant tracking, behavioral assessment, and production systems. The analysis found that no tested ATS keyword predicted sustained production after correction, while multi-system fusion improved predictive performance and exposed patterns no individual system could reveal.
| Unique skills parsed from ATS profiles | 8,181 |
|---|---|
| Keywords with enough data to test | 3,597 |
| Predicted sustained production after Bonferroni correction | 0 |
| Significantly anti-predictive keywords | 30 |
| Behavioral assessment AUC | 0.647 |
| Multi-system fusion AUC | 0.735 |
AUC measures ranking quality, not accuracy. To our knowledge this is the first production-scale decision-trace study of its kind. Source: arXiv:2604.19819, April 2026.
Where it started
I grew up in a small village in northern Pakistan with limited access to electricity, the internet, and technical resources. I began learning C# from a textbook, often writing programs on paper before I had regular access to a computer.
Years later, after 699 job rejections, I started Nodes because I had seen how badly rigid systems can misread people. The same instinct still drives my work: understand the system from first principles, then build a better one.
Selected writing
A production study of enterprise hiring records connected across three systems that had never been joined.
Explores limits in current silicon-centric AI systems and proposes a direction combining photonic or energy-based computation with specialized, dynamically activated subsystems.
Field notes, most days
Selected media
For enterprise work, visit Nodes. For research collaboration, technical discussion, speaking, or media, use the contact page.