Client Work

Built a multi-stage AI pipeline that turns raw survey data into evidence-based SWOT analyses. The application ingests CSV exports and automatically extracts themes, identifies patterns, and generates strategic insights with supporting evidence.

Built-in PII protection, checkpoint/resume so the pipeline won't choke on big datasets, and actionable "How Might We" questions on the way out. The workflow replaces manual spreadsheet review with repeatable API processing and consistent, data-driven outputs.

A UK apprenticeship education provider needed to turn one-line use case ideas into polished, classroom-ready content — fast. We built them a multi-agent workflow that handles it end-to-end. An editor submits an intent through the web UI; a fast clarifier asks targeted follow-up questions; nine ADK agents pick up from there — clarifying the query, running grounded research, drafting and polishing the outline, authoring the use case, and live-testing the walkthrough through the actual AI tool the lesson teaches.

The final draft auto-commits to a dated GitHub branch with full provenance. Hosted on AWS but reaches into the client's own S3 bucket for the customer knowledge base, so editorial output stays grounded in their authoritative data without ever leaving their cloud.

Used AI to develop assessment materials for UK apprenticeship programs focused on AI preparedness. Created multiple-choice quizzes, project-based assessments, and supporting resources including datasets and solution guides — all aligned with industry-relevant AI competencies.

Collaborated with a leading UK apprenticeship education provider to build assessments measuring learners' progress in AI skills from prompt engineering to data management. Applied AI-assisted research and prompt engineering throughout to create accurate, job-relevant evaluation materials.

Formulated online course curriculum teaching GitHub Copilot skills for a leading apprenticeship education provider. Courses cover practical Copilot workflows — code completion, chat, test generation, code explanation — with hands-on projects building real developer proficiency.

Built the course content and projects using prompt engineering and context engineering techniques. Applied systematic AI-assisted workflows to ensure technical accuracy while maintaining pedagogical effectiveness across the curriculum.

A caregiver training course on LearnWorlds for families and professionals supporting people with ALS. Five units across foundation knowledge, activities of daily living, quality of life, communication and support, and conclusion — built from expert-reviewed source material.

We handled LMS selection, curriculum architecture, and instructional design, plus the production pipeline that turns markdown source into LearnWorlds-ready HTML at scale. Working around a real LearnWorlds quirk: the OAuth API can create courses and units but can't update page content — so we built browser-console utilities to bridge the gap. Narration is generated through a custom ElevenLabs voice workflow (see Voice Production) so the team can iterate on script changes in minutes instead of booking studio time.

Built a custom ElevenLabs text-to-speech toolkit for a client whose voiceover budget was eating into their picture quality. AI voice freed up the budget for a professional film crew and full production studio for the live-action shoots.

The toolkit gives them what a traditional voice booking can't: rapid iteration on script changes — every revised line, regenerated in minutes, voice consistency maintained across long-form content via request stitching. The voice sounds entirely natural to viewers. The result: studio-quality video within budget, on schedule, with a film team they wouldn't otherwise have hired.

Built a RAG-based conversational AI for a healthcare client whose users needed answers grounded in their existing content — website pages, video transcripts, expert-reviewed source material — with proper source attribution and clear educational-only framing.

Streamlit frontend with native session-state auth (no third-party login libraries to babysit), FastAPI backend with vector search over the ingested knowledge base, deployed to managed cloud hosting. Every response cites the source URL it came from. Resilience built in — exponential-backoff retries, circuit breaker for the AI calls, friendly error pages when something downstream is down.

Created "Prompting 101" for a boutique branding and marketing agency. The reference guide covers prompt fundamentals through techniques including Chain-of-Thought, Few-Shot Learning, and XML-structured prompting.

Addresses agency service areas including brand development, marketing strategy, web and UX design, creative production, and research. Includes real client case studies and prompt templates for campaign briefs, creative briefs, and common audit workflows.

Built and host a privately branded Claude workspace for a small aerospace team. Two functional lanes — one for brand and business strategy, one for marketing and communications — each backed by a Claude-powered assistant with access to a shared knowledge base of strategy docs, brand guides, and past deliverables.

Authenticated through AWS Cognito, project files persisted in S3, served through CloudFront, and powered by the Claude Agent SDK on AgentCore Runtime. The team logs into a privately branded chat interface and gets two AI experts who already know their voice, their brand, and their history — without standing up cloud infrastructure of their own.

Ongoing coaching and mentorship for small businesses and educators on practical AI adoption. Topics include prompt engineering and context engineering strategies, AI workflows, using AI for research, and evaluating AI tooling options.

Sessions also cover AI risks and opportunities specific to each business, AI adoption strategies, developing AI policies, and building long-term AI strategy. Clients include small businesses across industries and the education company behind the Use Case Generator project.