James – AWS, Python, AI agent orchestration

Worked on the UK's largest pet microchip database. Built and deployed four production automation services that eliminated manual payment, support, and compliance processes, handling 16,000 workflows per month with zero manual intervention.

• Sole engineer on all four services end-to-end: architecture, build, deploy, client relationship, maintenance;
• Built payment-failure recovery (9k invoices/mo), missing-pet automation (500 reports/mo, saving 200 hrs staff time), regulated Transfer-of-Keepership workflow (1.5k/mo), and AI ticket triage (5k emails/mo);
• Integrated with Stripe, Zendesk, Microsoft Graph, Facebook, X/Twitter, SMS gateway, and upstream internal APIs;
• Built operator dashboards (Gradio + Next.js/TypeScript) for ops team visibility and audit;
• Migrated infrastructure from AWS to Railway, cutting hosting costs by 75%.

A multi-tenant generative AI image and video platform for two established UK fashion brands, enabling creative teams to generate lifestyle imagery, videos, and photorealistic renders from product photos and CAD designs.

• Designed and delivered the platform end-to-end, including system architecture, Next.js frontend, FastAPI backend, AWS infrastructure, fine-tuning, and dataset curation;
• Built six generative AI workflows, including a multi-model CAD-to-photorealistic rendering pipeline orchestrating multiple models in sequence;
• Curated and continuously refined a custom fine-tuning dataset for the flagship workflow, producing outputs rated best-in-category by prospective customers;
• Implemented a multi-tenant architecture with tenant isolation, user management, workflow controls, quotas, and asset segregation;
• Integrated multiple AI providers (OpenAI, Anthropic Claude, fal.ai) through an abstraction layer enabling model interchangeability across workflows;
• Delivered a production-ready frontend with authentication, generation history, favorites, and usage tracking.

An AI-powered digital avatar platform enabling real-time conversational interactions and text-to-video generation. Built core AI services responsible for speech synthesis, avatar orchestration, and multilingual text processing, powering low-latency avatar responses.

• Designed and shipped the streaming TTS deployment engine with sub-second latency for real-time avatar interaction;
• Extended the core Dealer orchestration service for avatar rendering — streaming Facesync (lip-sync from audio) and base-rig animation upgrades;
• Built a configurable multilingual text normalizer covering 14 languages (English, Arabic, Chinese, etc.);
• Maintained and stabilized the full SaaS Text-to-Video platform across all underlying services;
• Wrote performance-critical services in Rust; built ML integrations and supporting tooling in Python.

Research and engineering work in computer vision, machine learning, and image processing for UK national security and defence customers. Contributed to multiple classified projects spanning face recognition, adversarial ML, unmanned air system detection, and AI-generated text authorship verification.

• Deployed face-recognition services operating at tens-of-millions scale for national security and defence applications;
• Developed bespoke face-quality metrics that estimated per-probe false-alarm rates, providing operators with confidence and decision-support signals;
• Assessed physical and digital adversarial patches across multiple face-recognition algorithms in identity-avoiding and identity-targeting scenarios;
• Built a retraining pipeline for UAS (drone) detection with automated performance analysis and model-output visualization, significantly cutting per-experiment overhead;
• Dataset curation and systematic performance analysis across multiple model architectures;
• Researched authorship verification of AI-generated text.