AI systems I’ve designed and built. The common thread: encoding professional expertise into tools that stay contextually competent over time, rather than generating generic output from generic prompts.
TextbookAI
AI curriculum content generator for Sub-Saharan African markets, built for a global education publisher operating across multiple national curricula.
The challenge: producing locally-relevant curriculum content through traditional authoring is slow and expensive.
TextbookAI unifies manuscript and assessment generation into a single platform. Publishers upload a curriculum once and generate localised content, assessments, or both from the same source. A multi-agent pipeline with generator and evaluator always in separate API calls ensures quality gates at every stage. Human review is mandatory before downstream generation begins.
Results: Over 90% cost reduction.
CAPSPlanner
Weekly lesson planner for South African teachers working with the national CAPS curriculum.
The Annual Teaching Plan tells teachers what to teach each week. It doesn’t say how. CAPSPlanner automates that translation, generating a structured 5-day plan contextualised to each teacher’s actual classroom. A quintile 1 school with no projector gets different activities from an urban private school with a computer lab.
Teachers describe their school context once (province, quintile, class size, available resources, language profile). The system knows the CAPS ATP schedule and presents the correct topic each week with a generated plan tailored to that context. Plans can be regenerated by day, edited inline, and extended with worksheets, quizzes, homework sheets, rubrics, and parent communication notes.
Export to PDF, DOCX, and Google Classroom.
Forma
AI-powered instructional design platform. Not a content generator that produces flat text, but a system that encodes proven learning science as architectural constraints: backward design, Mayer’s multimedia principles, cognitive load theory, the 10–15 minute interaction rule.
The result is a course structure that a designer reviews and refines rather than builds from scratch. A full 6-module course that would normally take days of manual design takes under two hours of total user time.
Generator and evaluator are always separate API calls. Pedagogical rules are enforced through post-generation validation, not just prompt instructions. LLMs don’t reliably follow complex structural rules from prompts alone. A hard human checkpoint after blueprint generation means a wrong structure can’t multiply across modules.
Built for professional learning programmes across Sub-Saharan Africa and the Middle East, with a path to a broader product for instructional designers.
Export to Moodle (MBZ), DOCX, and PDF.
NiallOS
A working personal operating system for knowledge work, built over two years of daily use. Not a product you can buy yet, but a proof of concept for what expertise encoding looks like at the individual level.
NiallOS combines a structured knowledge database, project workspaces, document ingestion, verified extracts, reusable skills, semantic search, review queues, and agent workflows. It helps me manage consulting projects, MEd research, content strategy, and product development without losing the context that normally disappears into chats, folders, and people’s heads.
It is the prototype for what I build with clients at the organisational level: AI infrastructure that understands source material, workflow, professional judgement, and quality gates.
How expertise encoding works
- Map the work. Identify the decisions, documents, workflows, judgement calls, and quality gates that actually matter.
- Structure the knowledge. Turn project material, documents, contacts, decisions, and examples into a searchable knowledge layer.
- Encode the workflow. Build reusable skills, prompts, tools, and review processes that reflect how experts actually work.
- Add retrieval and evidence. Use search, embeddings, extracts, and source grounding so the system can find and cite the right context.
- Keep humans in the loop. Critical decisions, client outputs, and judgement-heavy work stay reviewable and auditable.
Work with me
I work with education organisations, professional services teams, and development partners who want AI systems that understand their context.
Typical work includes AI strategy and readiness assessments, expert knowledge capture, workflow mapping, AI product design, document-heavy knowledge systems, retrieval and synthesis workflows, and human-in-the-loop quality gates.
If your team’s most valuable knowledge lives in people’s heads, scattered documents, or repeated judgement calls, I can help you turn it into AI infrastructure that lasts. Read more about expertise encoding, see how to work with me, or get in touch.