Learning by doing
In a fast-moving AI and tech landscape, staying effective demands hands-on experimentation. These self-directed projects reflect how I continuously build, test, and adapt outside of formal client engagements.
How do you know an AI system is good enough to trust?
Designing evaluations with synthetic data and adversarial inputs to test whether model outputs hold up. Every public-sector AI rollout needs this.
Where does AI save time, and where does it still need a person?
TrAIvel Concierge, a group-travel planner that grounds its suggestions in real data and was tested with automated and human evaluation. And an n8n workflow that turns unstructured emails into structured records, with human review built in.
Can an AI coding agent rebuild a website safely?
This site, redesigned with Claude Code: brand brief, clean-up, and design, with security rules set before any code changed.
What does public data show that headlines miss?
Data Driven Pakistan, a four-part series built on public data, and an analysis of funding under BEAD, the US federal broadband program.
