Sam S.

The London digital agency needed to serve more clients without adding headcount. I designed AI workflows that handle the mechanical work so humans stay in control of what matters.

CHALLENGE

The agency managed WordPress content, customer support, and marketing campaigns manually across multiple clients. Response times averaged 4+ hours, content production bottlenecked on senior staff, and quality varied with workload.

APPROACH

I facilitated workflow mapping with the agency director and team to identify friction points — repetitive tasks, scattered validation, wait times between steps. I prioritized automations by measurable ROI: starting with WordPress (highest time consumption), validating the approach, then expanding to support and marketing.

SYSTEM DESIGN

Designed and architected a multi-agent system with domain-specific AI agents (WordPress, Support, Marketing), each with defined roles, confidence scoring, and human escalation logic. Core design principles: preview-before-publish, transparent AI behavior, human override at every step.

Technical architecture: FastAPI + Redis Streams event orchestration, Pydantic state machines, multi-channel integration (web chat, email, WhatsApp). AI governance aligned with ISO/IEC 42001 and NIST AI RMF frameworks. Observability through Langfuse and Logfire.

RESULTS

Beyond interface design, this project demonstrates designing how AI systems behave: identity, autonomy boundaries, trust signals, and escalation patterns.

Repetitive practice is the hardest engagement problem. I designed and shipped a dual-experience platform where skill-building feels like play and stakeholder analytics drive timely action. Research to production in 16 weeks.

RESEARCH

Structured interviews with teachers, parents, and private tutors to understand how they teach and track conjugation practice. Four insights shaped every decision: no time for proactive follow-up, existing tools are frustrating, progress tracking is buried, every teacher has unique styles.

Competitive analysis across 5 tools revealed a market gap: engagement OR educational rigor, never both. No tool combined gamification with stakeholder analytics.

DESIGN

Dual-experience architecture serving two audiences with opposing needs:

- Learner app (mobile-first): white-hat gamification using Octalysis framework — mastery, curiosity, creativity. Spaced repetition algorithm personalizing review intervals. Self-guided progression with autonomous learning path.

- Stakeholder dashboard (responsive): real-time analytics surfacing patterns before they become problems. Granular difficulty tracking enabling timely teacher/parent intervention.

AI-ASSISTED WORKFLOW

- LLM content generation: 80% reduction in exercise creation time

- AI UI generation tools: rapid prototyping and component exploration

- Stable Diffusion + ComfyUI: batch production of consistent character assets

- Code assistants: 60% velocity gains on full-stack implementation (FastAPI, React, PostgreSQL)

- QA automation: Playwright MCP integration for systematic testing

All AI workflows documented as transferable playbooks with prompt versioning and human review gates.

RESULTS

- Production MVP validated through user testing with teachers, parents, and K-12 learners

- WCAG 2.2 AA compliant across all interfaces

- Demonstrates end-to-end capability: research, UX/UI, gamification design, data visualization, full-stack development, AI-assisted workflows, accessibility, and QA

A platform trusted by ministries worldwide needed a modern authoring experience. I led the redesign from AI-assisted workflows to familiar interaction patterns, validated across 400+ users and three markets.

CONTEXT

TAO is a 10-year-old assessment platform deployed in 70+ countries. The authoring experience needed modernization: integrating AI-assisted workflows, making complex professional tools feel familiar, and validating across diverse markets before committing engineering resources.

APPROACH

Applied an evolved Triple Diamond methodology: strategic framing, user discovery, implementation leadership. Three departments had competing priorities — Marketing needed campaign material, Engineering pushed for complete rebuild, Product required business model validation. I built a "Preview Alpha" prototype enabling simultaneous marketing demos, B2C vs B2G hypothesis testing, and engineering constraint mapping.

RESEARCH

Led international user research: 400+ participants across EU, Japan, and US markets. 30+ in-depth interviews. Mixed methods — surveys, workshops, usability testing. Research validated strategic pivot from B2C to B2G, informing executive investment decisions.

DESIGN

Applied IDEO DVF framework (Desirability-Viability-Feasibility) for systematic trade-off analysis at each decision point. Co-creation workshops with teachers ensuring advocacy for end-user needs.

RESULTS