Simon Cruise

I helped develop an AI Agent repository platform for a leading multinational consumer products company. I participated in daily planning sessions, refining requirements, and translating them into technical specifications. I designed and built scalable applications using FastAPI, React, NextJS, and PostgreSQL, while supporting CI/CD pipelines for reliable deployments.

Working in an Agile environment, I collaborated closely with the Product Manager to manage the backlog, estimate user stories, and prioritize features aligned with a 2025 roadmap. I wrote clean, maintainable code, conducted code reviews, and troubleshot issues to ensure application stability. I also integrated OpenAI API to enhance AI capabilities, driving innovation and delivering future-ready solutions that support the client’s market leadership.

I was contracted to add an organization tier to a live Next.js/Supabase SaaS that only modeled flat teams: one business entity owning many teams, with org admins managing membership across all of them. The client still merges upstream releases of the starter it is built on, so every change had to be additive—new schema files, new migrations, new routes, no edits to inherited code.

Authorization was the hard part. Organization admins needed to reach into every team in their org without a membership row per team, and checking that per request meant recursive queries. I resolved permissions at token-mint time instead, using Supabase's access-token hook to union role, owner-grant, and per-app permissions across a user's accounts into a JWT map of account id to permission keys with namespace wildcards, then backed row-level security with a single security-definer helper that settles an org admin's reach over any team in one join.

I shipped 11 SECURITY DEFINER RPCs, 50 migrations, 274 Playwright tests, and RLS specs that sign in as real users so policies run against real JWTs.

Built a full-stack football match prediction platform: ingestion, feature engineering, training, calibration, inference, and a public web app. It pulls fixtures, standings, and match statistics from a sports data provider into a relational database, then derives Elo ratings, venue-specific form, pre-match standings snapshots, and recency-weighted shot/xG aggregates around 200 features per fixture.

Four model pipelines cover domestic leagues, cups, continental club cups, and internationals, whose dynamics differ sharply. Each pair of XGBoost Poisson goal regressors (Dixon-Coles low-score correction) with a multinomial home/draw/away classifier, using Optuna search, temporal splits, and Bayesian-smoothed priors. A matrix runner trains competing feature profiles side by side, so promotions are evidence-driven.

A probability that says "73%" has to win 73% of the time, so every artifact ships with isotonic calibrators fitted on the holdout, closing an 8-12pp overconfidence gap at the top end. Production runs as independent cron jobs with monitoring, including a coverage check for silent failures where a pipeline "succeeds" while producing nothing. The front end is a Next.js 16/React 19 app serving daily tips and accuracy records.

I built and shipped a mobile-first Progressive Web App for a new B2C personal jewelry insurance product distributed at point-of-sale in jewelry and watch retailers. The onboarding flow captures customer details, plus receipt/item photos (with basic clarity checks), and then starts a Stripe-backed 7-day free trial that auto-converts to an annual cover unless cancelled.

I owned the payments and policy lifecycle end-to-end, including payment method holds, trial-to-paid conversions, renewals, and robust Stripe webhook handling for success/failure states with retries, error handling, and an audit-friendly log trail. I implemented secure photo storage with signed access and optimized the experience for fast in-store signups.

I automated policy artifacts and comms: sequential policy numbering, Day‑0 and Day‑7 policy schedule PDFs, and transactional emails via Postmark (confirmation, conversion, renewal reminder) with email event logging. I also delivered retailer conversion analytics, internal support/admin search tools, HubSpot sync for agent visibility, and a daily policy/payment JSON export for the NetSuite team.

Built an end-to-end engine that converts raw Figma .fig files into responsive, production-ready HTML—no Figma API or plugin. The core reverse-engineers Figma's undocumented binary format (kiwi encoding over zstd/zlib) into a normalized document tree, then rebuilds the semantic structure from absolutely positioned design layers.

The hardest part was component instance expansion: Figma stores overrides as sparse, GUID-keyed diffs whose keys resolve to different base nodes across instances. I built a seven-layer matching algorithm—override keys, structural DFS, and geometric heuristics—to reconstruct the full rendered tree, plus a vector pipeline that emits SVG from raw path blobs and font glyph outlines. Blend modes and masks that CSS can't express are pre-composited via headless Chromium at render time.

A vision-LLM layer then classifies page sections from screenshots. It rewrites absolute markup into responsive Tailwind, guided by measured geometry so the model never guesses facts the source file already knows. Shipped as both a Typer CLI and a SaaS: FastAPI/SQLAlchemy back end with a job worker, Stripe billing, and Alembic migrations, plus a React 18, TypeScript, and Vite front end.

I helped maintain around a dozen sites while working with many internal and external stakeholders to manage the corporation's online presence and its umbrella companies and organizations. Because of the large size of the organization, I was required to adapt daily to meet the client's needs on various technology stacks and frameworks.

I developed and maintained a Node.js back end powering a content management system (CMS) for a major real estate company. The CMS, built with Node.js and Vite, enabled content editors to efficiently manage property listings and site content. I designed and implemented APIs that served the main website, ensuring seamless, high-performance data delivery to support a dynamic user experience.

I contributed to requirement gathering, technical planning, and Agile development processes. Also, I focused on writing clean, maintainable code, optimizing back-end performance, and ensuring the CMS and API remained stable and scalable. This role demanded rapid problem-solving and adaptability to evolving business needs, delivering solutions that enhanced the company’s digital presence and customer engagement.

I was the tech lead for a gaming API used by the BBC to enable high-score tables, user storage, multiplayer features, and authentication within online applications. The API was used by various applications, from games and online quizzes to storing users' favorite recipes.

I created an API to provide quiz functionality to online applications. The quizzes cover pop culture, general knowledge, and specialty subjects, with some being oriented to helping school-age children learn and practice for their end-of-year exams. The API had to support progression, feedback, and authentication and provided support for hundreds of requests per second.

I was involved with this project from the UX to the content generation stage and client training for the content management system (CMS). I worked alongside the client, creatives, and project management team to deliver a forward-looking, easy-to-navigate web app packed with helpful content.