Chris C.

AI Integration Developer | Claude API, Workflow Automation, WordPress

AI integration on systems that already exist and have to keep working. I build Claude API features into live codebases: chat that remembers across devices, workflows that run unattended, and safety gates that hold when someone tries to break them. I'm Chris, a senior full-stack developer based in South Africa (GMT+2), with strong overlap on EU and US hours. I think through architecture, edge cases, and long-term maintainability. Clients have called me "a thinking developer who anticipates needs" and described my work as "exceptional precision and judgment". Most AI work fails at the same place. Not the model call, which is easy, but everything around it: what happens on a bad answer, who is allowed to see what, what you can show an auditor six months later. That is the part I build. What I do (ranked by recent revenue and market fit): • AI integration into existing products. I built a WordPress plugin giving an AI companion cross-device memory per logged-in member, rolling summarisation so long conversations stay inside the token budget, and a content-safety gate screening both what the user sends and what the AI replies, across fourteen compliance categories, with a safe fallback and a logged, auditable record of every blocked message. Red-teamed against six attack routes including non-English, encoded text, fictional framing, multi-turn build-up and prompt injection, then promoted to live and verified there (5.0 stars, "a complex, compliance-heavy piece of work, and he delivered exactly"). Built on the existing stack without modifying the theme, the plugins or the store. If your AI feature has to survive a compliance review, this is the work. • AI workflow and automation systems. I design and run agentic systems, not demos. My own practice runs on one I built: fourteen Claude Code skills with a shared context layer, a Model Context Protocol server exposing structured tools the model calls directly, a Postgres audit trail recording every run, approval gates so nothing writes without a human diff, and a reconciliation loop that has to match an external financial statement to the cent. It is proprietary so I cannot hand over the code, but I will walk you through it live. If you want AI put to work inside a business rather than bolted on the side, this is the shape of it. • Custom JavaScript logic, calculators, and interactive tools. A $2,000 custom financial platform with persona routing, JS calculators, and LiteSpeed tuning (5.0 stars, "restructuring site architecture, committed to quality"). Three live demos you can click and try are in my portfolio: a real-time pricing engine, a multi-step loan eligibility calculator, and a subscription pass with QR redemption built on Stripe. • Full-stack feature work in existing codebases. A 152-hour senior engagement on a vanilla PHP, JavaScript and PostgreSQL codebase ($5,320): features, experiment plumbing, integrations. I own problems inside custom themes and frameworkless stacks, not just configure plugins. • WordPress, Stripe and the rescue work. Theme migrations and PHP upgrades (5.0 stars, "excellent from start to finish... delivered on a tight timeline"), Stripe discount logic on a live payments page (5.0 stars, "fast and effective, most efficient solution"), server-side purchase tracking, and CMS-editable custom logic so you can manage the values yourself. Staging first on WP Engine or Flywheel, full backup, automated browser tests before go-live. How I work: • Day 1 to 2: read the brief twice, ask the sharp questions, write a short scope doc with milestones and a flat estimate. On anything complex the first milestone is a paid audit, so we scope the build from what I actually find rather than from a guess. • Week 1: first working increment on staging or a preview URL. Never silence. • Ongoing: short async updates with screenshots, blockers flagged within a working day, clean Git history. • At handover: a README another developer can pick up cold. One thing I am straight about on AI work, because it matters. I build and test layered guardrails: prompt-level rules, a moderation pass on input and output, a hard block with a safe fallback, and a log of every blocked event. What no one can honestly promise is that a probabilistic model will never once produce something off. What I can promise is a tested, auditable gate mapped to your rules, with the log as the evidence trail. Clients in regulated categories tell me that is the answer they were looking for. Not a fit if you need a same-day patch on a site you cannot grant access to, 24/7 chat-style support, model training or fine-tuning from scratch, a 6-month framework build in Laravel or Symfony, or someone to manage subcontractors. I work hands-on, solo, on bounded scopes, and I take on fewer, larger projects rather than many small ones. If you need someone who reads the brief carefully, owns the problem, and leaves the codebase in a better state, that is the work I do best.