LLMs, RAG, agents and fine-tuning, shipped inside real applications.
Most AI features never make it into the product. They stall as a prototype, or a contractor hands over a model nobody can actually use. I build the ones that ship.
Retrieval systems over your own data, AI agents and fine-tuned models, delivered inside a real application your customers use on day one. The model and the product from one person.
I work with Australian businesses, and the agencies that serve them. SaaS startups, and established companies adding AI to products that already run. Over a decade of Laravel, PHP and API work behind every delivery. Plus the server management and security to keep it running.
Open source models on Hugging Face with tens of thousands of downloads. 100% job success, over eight years on Upwork, 1,700+ hours with long-term retained clients. Director of DreamFast Solutions Limited, a Hong Kong registered company.
Send me a message. I work Australian business hours and reply fast, and I will tell you honestly whether AI is the right tool for your problem.
Thirteen different groups abliterated the same AI model, Google’s Gemma4-E2B. Every single one removed the safety filters. That part is not interesting any more. What is interesting is how much collateral damage each technique caused along the way, and how many of the creators’ capability claims survived an independent measurement. The KL divergence spread between the best and worst variant is 58.7x, the largest I have ever seen in this project. And one creator’s “near-zero divergence” claim turned out to be 187 times lower than reality.
Five different groups abliterated the same AI model. When I ran the maths benchmarks, their scores ranged from 27.5% to 75.1%. That is a 47.6 percentage point gap. It looks like some techniques made the model way better at maths and others broke it. But when I dug into why, it turned out nobody got smarter or dumber. The abliteration just changed how long they think before answering. The real scores were all within 2.8 percentage points of each other.
HauhauCS publishes uncensored LLM models on HuggingFace with 5 million combined monthly downloads. Every model card claims “private methods and tools.” I recovered the deleted source code from PyPI’s CDN. It is a fork of the open source Heretic abliteration tool, refactored and relicensed without attribution.
Fine-tuning and abliteration of open source language models, with tens of thousands of downloads on Hugging Face. Model quantisation and optimisation for consumer hardware, plus dataset preparation and curation.
Agentic AI systems and RAG pipelines, integration with LLM provider APIs, and prompt engineering and evaluation for predictable behaviour in production.
Modern web application development using HTML5, CSS frameworks like Bootstrap and TailwindCSS, along with SCSS and LESS for styling.
JavaScript expertise includes jQuery, AJAX, and vanilla JS, with frontend frameworks such as Vue.js, React, Next.js, and Nuxt.js. Backend work spans PHP and Laravel for API development.
Database experience includes MySQL, MariaDB, PostgreSQL, and SQLite. Additional skills include responsive design, cross-browser compatibility, and Git version control.
Automated build and deployment pipelines using GitHub Actions, GitLab CI and shell scripting. Staging and production environments with automated testing, zero-downtime deploys and rollback strategies.
Containerisation with Docker for reproducible builds and isolated environments, so deployments are predictable rather than hand-tuned.
Server management across Debian, Ubuntu, CentOS/Almalinux, and BSD operating systems, including WHM/cPanel setup. Web server expertise covers Apache2, NGINX, and Caddy with load balancing and reverse proxy configurations.
Self-hosted email servers using Postfix, Dovecot, and SpamAssassin with ClamAV, implementing spam protection through DKIM, SPF, and OpenSRS.
Additional capabilities include shell scripting, firewall configurations with ipset and blacklists, VPN setup with OpenVPN and WireGuard, disaster recovery, backup strategies, and CDN/DNS services.
Penetration testing services including scanning local networks and software for vulnerabilities, along with remote vulnerability scanning.
Post-incident services include cleaning and restoring hacked or malware-infected websites, and securing servers after an attack.
Experience with WordPress and Shopify, including custom theme design, modifications, and custom plugin development.
SEO optimisation implemented to ensure sites rank well in search engines.
For over a decade I've shipped production software for clients across different industries. The work below is a mix of AI systems and the applications around them.
Integrated LLM-based features into existing Laravel applications for clients, including AI-assisted content generation, intelligent search and automated data processing. Built RAG pipelines for domain-specific knowledge retrieval, and fine-tuned open-source models for client use cases. Also release my own open-source models on HuggingFace.
Architected and built a multi-tenant SaaS application from the ground up using Laravel and Vue.js. Designed the API layer, tenant isolation, subscription billing integration and role-based access control. The platform scaled to serve thousands of active users across multiple organisations.
Took over a large legacy PHP codebase for an established business and incrementally migrated it to a modern Laravel architecture. This included restructuring the database layer, building a RESTful API, implementing automated testing and deploying with zero downtime to ensure business continuity throughout the transition.
Developed a custom Laravel API backend for an e-commerce platform, integrating with third-party payment gateways, inventory management systems and shipping providers. Focused on security hardening, PCI compliance considerations and performance optimisation under high traffic loads.
Built internal tools and workflow automation systems for small and medium businesses. Everything from CRM-style platforms to booking systems to data dashboards. These projects typically involve close collaboration with non-technical stakeholders to translate business requirements into clean, maintainable software.
Designed and managed production server environments for multiple clients, including load-balanced configurations, automated deployment pipelines, firewall management, email server setup with full DKIM/SPF/DMARC compliance and ongoing monitoring. Provided security audits and vulnerability assessments across all environments.
Set up automated build and deployment pipelines using GitHub Actions, GitLab CI and shell scripting. Configured staging and production environments with automated testing, zero-downtime deployments and rollback strategies. Just reliable pipelines that don't need babysitting.
Built and customised Shopify stores for clients across retail, services and hospitality. This includes custom theme development, app integrations, payment and shipping configuration, SEO and performance tuning. Most of these projects involved taking an existing store that wasn't working well and rebuilding it into something the client could actually manage themselves.
