AI in Practice
Before it became standard practice. These are specific things I have done, with specific outcomes — not a list of tools I have heard of.
"'The most useful thing about AI in design work is not the output. It is what it forces you to articulate to get there.'
On using AI as a thinking partner, not just a production toolAI synthesis is fast and imperfect. The discipline comes in knowing which imperfections matter and which ones do not — and in structuring the inputs well enough that the outputs are actually useful. That is a design skill, not a tech skill.
Specific Work
2023 · Hackajob
800 candidate surveys synthesised using Dolphin AI to surface hiring behaviour patterns at scale. This was before AI synthesis was a common design workflow.
2023 · Hackajob
Board of Innovation AI frameworks applied to opportunity mapping and big bet confidence scoring — structured the product direction for Engage's second phase.
2023 · Hackajob
Future state scenario mapping including a 2026 hypothesis called "Scouts" — an intelligent hiring guide concept that has since mapped onto what Hackajob now calls AI agent recruiters.
2024 · Deel
Performance digest and AI development plan generation explored at Deel. Thinking reached development post-departure and has since influenced longer-term roadmap direction.
2024 · Deel
Conversational AI prototypes built to explore decision support interfaces for hiring managers — mapping intelligence onto the moment when managers need to act, not after the fact.
2024–Present · Matcheros
AI-assisted job description generation and CV parsing workflows designed and shipped — making it practical for Africa-first hiring teams to move fast without sacrificing quality.
The intelligence tool concept I developed at Hackajob — an agentic hiring guide that surfaced the right candidate at the right moment — mapped onto what Hackajob has since built with their AI agent recruiters. The hypothesis was right two years early.
Conversational Decision Support
Built conversational AI prototypes at Deel to explore a different interaction model for performance data — one where the system surfaces the insight rather than waiting for a manager to find it. The question was whether intelligence could arrive at the moment of decision, not after it.
'The AI synthesis saved weeks of manual work and surfaced patterns we would never have found manually.'
CS Team Lead · Hackajob · on the 800-survey synthesis workflow