Problem
Case study 01 · Revise Right
Applied AI for exam revision
I designed, shipped, and operate a live revision platform that turns exam-board scope into notes, quizzes, flashcards, mind maps, study packs, and AI support students can actually use.
Open the live product- 11
- live subjects
- 54
- exam-board pathways
- 410
- published topics
01Exam board + level
→
02Scoped source context
→
03Provider routing
→
04Automated QA + remediation
→
05Learner tools
System
One thread from syllabus to study tool.
Public topic routes feed a multi-provider generation layer, content checks, and Firebase-backed tools for notes, quizzes, flashcards, diagrams, and papers.My role
Founder, product designer, and builder.
I own the product logic, model behaviour, evaluation approach, interface, content operations, and the decisions required to get it into production.Difficult constraints
Specificity, consistency, latency, and trust.
The system has to stay aligned across boards, levels, and subjects while keeping generated work fast, legible, and useful to a student under pressure.How I tested it
Representative cases, QA checks, then failure review.
I test across subjects and exam pathways, compare outputs against the intended scope, run automated content checks, and review edge cases through the full learner flow.In production
Real scope, not a prototype.
The product now serves 11 subjects, 54 revision pathways, and 410 published topics, with public notes and past-paper hubs leading into the wider AI study stack.Also building
Grounded Magazine
I created this independent magazine end to end—from its identity and information architecture to the website, CMS, editorial workflow, and publishing cadence.
Confidential client work
I am shaping an early-stage product across naming, positioning, product logic, operational workflows, and practical AI automation.
How I work
Principles I build by
AI has to survive real use
A model is only part of the product. Retrieval, context, evaluation, feedback, transparent uncertainty, and clear boundaries determine whether it remains useful after the demo.
The interface should protect the thinking
The best tools reduce cognitive load without flattening the work—especially in learning, research, and writing, where the thinking is the point.
People should be able to own their tools
I am drawn to local models, open software, and small publishing infrastructure that people can inspect, adapt, and control.
Systems behind the work
What I have worked with—and why
Model layer
OpenAI and Claude power generation and routing in Revise Right; the product work is deciding which context, boundaries, and checks make those models dependable.
Product + state
React and Firebase support the live learner interface, saved resources, and the student workflows around the model—not just the prompt itself.
Evaluation + QA
Automated content checks sit inside the production path. OpenAI Evals informs how I think about repeatable cases, graders, and failure analysis.
Canvas interfaces
Excalidraw and Canvas UI shape how I approach interactive diagrams and mind-map tools without sacrificing usable HTML.
Local agents
Ollama and Hermes Agent shape my experiments with local inference, persistent context, tool use, and reusable skills.
How I got here
I build technology, but I start with people.
I came to software through publishing, branding, and education rather than a traditional engineering path. That keeps me focused on both layers: what a product is doing under the hood, and whether someone can understand, trust, and use it. I learn by shipping, reading source, testing edge cases, and changing the solution when the evidence changes. I keep rebuilding until the logic and the language agree. Away from the screen, I am usually walking Winston—objectively the best dog in the world.
Hiring for applied AI—or building something difficult?
Me + Winston
The best dog in the world
Modes: