London-based digital product builder

Ferdinand Macagba

AI-assisted application developer with product judgment, responsible AI discipline and live shipped work.

I research the problem, define requirements, specify the workflow and direct every revision. AI assists implementation. I own the product decisions, review and release.

From Operations to AI-Assisted Application Development

Over a decade in hospitality and remote operations, from e-commerce fulfilment to distributed teams, taught me what organisations need from software: tools that work reliably and workflows built for the people who use them every day. Since 2023 I have been building that software.

My Development Process

Five stages, one human approval gate, and every failure loops back to the stage that caused it.

How the work flows: the build loop
  1. PRD / Spec

  2. AI-Assisted Build

    AI implements

  3. Human Review

    ◆ approval gate

    reject → revise build

  4. approve
  5. Test & Verify

    spec gap → back to PRD

  6. pass
  7. Ship

    bug fix → back to AI Build

  8. iterate → next PRD / Spec
Human-led AI-assisted

The loop in practice: BytesCRM

AI-Powered CRM with Embeddable Chatbot

Problem and research
Small businesses lose website visitors without capturing them as leads, and the tools that exist fail solo founders: per-seat pricing, no website lead capture, and chatbot, CRM and billing split across three products.
Requirements (PRD)
Embeddable chatbot widget (single script tag), lead inbox with overflow-and-unlock, CRM pipeline, autoresponder sequences, Stripe tiered billing with pay-per-lead unlocks, and role-based access (User, Admin, Developer).
Architecture decision
Multi-tenant SaaS with isolated data, chatbot configuration and billing per business account. React 18 front end, Base44 serverless backend, Stripe, Resend, and OpenRouter with Claude and Gemini behind a chatbot that ships as a standalone embeddable widget.
Show 4 more: AI-assisted implementation, Responsible AI, Iteration, MVP
AI-assisted implementation
Each module scaffolded with AI assistance, output reviewed, gaps identified, corrections specified, then retested. Human sign-off at each module before moving on. AI provider cost monitored via a built-in pricing governance dashboard.
Responsible AI
Chatbot conversations are rule-scoped: AI cannot take actions outside its defined flow. Role-based access enforced throughout (User, Admin, Developer). Stripe environment switching (Test vs Live) requires explicit developer-level action with full audit logging. Overflow lead expiry emails give business owners 48h and 24h warnings before leads are permanently removed.
Iteration
Five real pain points drove post-launch additions: losing website visitors (added chatbot widget); cost barrier (free tier with overflow model); attribution blind spots (UTM, Google Ads click ID, Meta Pixel tracking); onboarding friction (guided setup, under 10 minutes to first lead); and account recovery (structured deletion cooldown and re-registration flow).
MVP
Chatbot widget + lead inbox + CRM pipeline + Stripe billing deployed. Tiered plans (Free, Starter, Pro, Business, Enterprise) live. Smart Chatbot AI add-on and message top-up packs in the roadmap.

Featured Projects

Bytes is my own brand, so the Bytes-named products are mine rather than client work.

Proof of process

A Shopify catalogue that stocks itself from visitor demand; a product goes live seconds after the first click on it.

146 tests in real workerd
The Vitest suite runs inside the real Cloudflare Workers runtime with no network mocks, not in a Node simulation.
The outage that hardened the gate
An LLM outage let kitchen tools through the quality gates. A worker-level pet-relevance guard now keeps the store stocked with genuine pet products with or without the LLM.
A guard test that fails the build
Only one module may touch D1; a guard test enforces that tenant boundary and fails the build if another module crosses it.

Key decision. Everything runs through Shopify App Proxy, so if PawScout goes down the store falls back to plain Shopify search with no errors shown.

Shopify Liquid theme and App Proxy · Hono on Cloudflare Workers · D1 · Vitest

  1. BytesAI Learn homepage: curated free AI courses with a searchable catalogue

    BytesAI Learn

    A hand-curated catalogue of 100+ free AI courses from 20+ providers, searchable without signing up.

    Next.js 16 on Cloudflare Workers via OpenNext · D1, KV and R2 · Gemini recommendation chatbot

  2. BytesCRM landing page showing the live lead-pipeline dashboard

    BytesCRM

    One script tag captures website leads into a CRM pipeline; businesses start free and pay per lead as they grow.

    React 18 · shadcn/ui · Base44 serverless (Deno) · Stripe · Resend · OpenRouter with Claude and Gemini

  3. Pivot Parlor booking platform landing page with a sample booking widget

    Pivot Parlor

    Every barbershop gets its own AI-generated booking site; customers book as guests, owners run staff, queues and revenue.

    Remix on Cloudflare Pages · Hono API on Workers · D1, KV, R2 and Queues · Stripe Connect · Turborepo

  4. Shikaku: Puzzle Quest game menu with play and battle-arena options

    Shikaku: Puzzle Quest

    100 hand-crafted logic puzzles and live 1v1 ELO battles in a PWA that plays offline.

    Vanilla HTML, CSS and JavaScript · HTML5 Canvas · Cloudflare Pages, Workers and KV

  5. Oak Hill Park Cafe redesign hero with 5-star hygiene certificate and soft play info

    Oak Hill Park Cafe

    Real-time party-slot booking for a Barnet family cafe, proposed as a rebuild timed to their new 5-star hygiene rating.

    Dependency-free static front end · Cloudflare Pages Functions · D1 · Resend

Show the toolkit behind these projects
AI-assisted development
Next.js / React project direction, TypeScript-based app workflows, Node.js / API feature planning, REST API integration concepts, Tailwind CSS UI iteration, AI-assisted debugging and review, VS Code
Data and storage
SQLite and Cloudflare D1, Database design, Cloudflare Workers
Platforms
Shopify (custom themes), GitHub, Cloudflare, Google Cloud Platform (learning)
AI tools
Claude, ChatGPT, Gemini, GitHub Copilot, Ollama, Cursor, Base44, AntiGravity

How I Build Safe Applications

Six controls run on every project. Each one below names where it is already live.

  1. Input Validation & Data Quality

    Format, type and length are validated on every input, and external content is untrusted by default, before anything reaches an AI or backend system.

    Live in: Active Paw ghost-card quality gates · BytesCRM contact imports

  2. Prompt Injection Awareness

    Separate trusted system instructions from user-supplied content. Restrict what AI can access. Test adversarial inputs during development, not just happy-path scenarios.

    Live in: PawScout LLM quality gates · BytesAI Learn chatbot and AI article generation

  3. Human Approval Gates

    AI-generated content requires human review before publishing. Consequential actions (releasing features, publishing pages, triggering automations) require explicit human sign-off.

    Live in: Pivot Parlor: staff reviews AI pages before going live

  4. Output Verification

    Check AI outputs against requirements. Cross-check factual claims. Confident-sounding output is not a substitute for verification; hallucination is always a risk, treated as such.

    Live in: PRD critique workflow · Case study content review

  5. Permission Controls & Guardrails

    Least privilege throughout: no AI system holds more authority than its task needs, and structured outputs, rate limits and error handling contain failures.

    Live in: BytesCRM role-based access (User, Admin, Developer) · Stripe environment gating

  6. Context Engineering

    Purpose, requirements, decisions and task lists persist across sessions, and a handover document restarts the work before context limits let errors compound.

    Live in: This portfolio, built and maintained using this exact practice

Education & Certifications

  1. April 2026

  2. April 2026

  3. February 2026

  4. Bachelor of Science in Electronics and Communications Engineering (opens in a new tab)

    Saint Louis University, Philippines

    UK Equivalent: Foundation Degree / HND Level (UK ENIC Statement of Comparability, April 2023)

    1987

Work Experience

  1. 2023 – Present

    London, UK

    AI-Assisted Application Developer

    Independent Projects

    • Shipped 5 live applications and 1 client proposal across SaaS, e-commerce, EdTech and games
  2. July 2019 – December 2022

    London, UK

    Remote Operations Manager / E-Commerce Specialist

    Bytes Enterprises Ltd

    • Managed customer communications and order operations remotely for a high-volume e-commerce business
    • Coordinated 10+ suppliers and logistics partners; reduced order-processing time by 20% while maintaining 99% inventory accuracy
  3. April 2010 – July 2019

    London, UK

    Night Receptionist / Central Reservations Agent

    Radisson Edwardian Hotels

    • Managed front-desk and reservations operations independently; processed complex bookings with 100% accuracy on client details
    • Trained new staff on reservation procedures and customer-service standards

Available for London and remote roles

Open to full-time, contract and freelance work, on-site, hybrid or remote. Send the role or the problem and I will reply within one working day.