I'm a .NET and Azure architect who builds production systems, APIs, cloud platforms, automation, and data pipelines. Lately most of that work involves wiring LLMs into real products and keeping them on a leash: validation, guardrails, and cost control so the AI helps without going off the rails.
I run Soltech Consulting Ltd, where I take on architecture and hands-on engineering for teams that need something shipped rather than a deck about shipping it. In practice that means owning the whole path from design, build, CI/CD, observability, and the operational corners everyone forgets until production. That work has included enterprise systems for clients like Microsoft, Rolex, Jaguar Land Rover, Coca-Cola, Bowers & Wilkins, ARAMCO, and SABIC.
Most of my open work sits at the intersection of audio and AI: harmonic analysis, recommendation, metadata enrichment, and staged LLM orchestration. It comes from the other half of my life, I DJ as Changsta across House, Drum & Bass, Breakbeat, UK Garage, UK Bass, Funk, and Hip Hop, which is where most of these problems start.
What I'm digging into right now:
- AI-assisted engineering workflows
- LLM orchestration and guardrails
- Automation pipelines
- Recommendation systems
- Azure-native architectures
- Developer tooling
- Music-tech systems
Available for fractional architecture and AI-integration work. If your team needs an experienced pair of hands on .NET, Azure, or shipping LLM features safely, let's talk or reach me on LinkedIn.
Languages C# · Python · SQL · JavaScript · TypeScript
Frameworks .NET 10 · ASP.NET Core · Web API · Entity Framework Core · React
Cloud Azure App Services · Azure Functions · Azure Storage · Azure SQL · Azure DevOps
AI / Automation OpenAI · Claude · Gemini · Groq · Mistral · OpenRouter · Ollama
Infrastructure GitHub Actions · Bicep · Docker · Cloudflare Workers · Cloudflare Pages
Observability OpenTelemetry · Azure Monitor · Application Insights
Data / Messaging RSS · JSON APIs · Discord Bots · SoundCloud RSS ingestion
Tooling Git · VS Code · Claude Code · Rekordbox · Mixed In Key
| Project | What it does |
|---|---|
| MixLab | AI-assisted DJ set generation that reads a Rekordbox collection and builds structured mix concepts from harmonic compatibility, energy flow, and enriched metadata. Uses staged LLM orchestration rather than one oversized prompt, so each step stays cheap and checkable. |
| TuneFinder | Music discovery pipeline that watches new releases across platforms, scores them against how I actually mix, and posts a curated weekly report to Discord. |
| Rekordbox Metadata Enrichment | Fills gaps in Rekordbox libraries using MusicBrainz and Discogs, with confidence scoring, caching, and LLM-assisted disambiguation for the ambiguous cases. |
| SoundCloud AI Mix Recommender API | The .NET API behind changsta.com. Pairs deterministic validation with AI reasoning to recommend mixes from structured catalogue metadata — the AI suggests, the rules decide. |
- Production-first: I design for the day it's live, not the demo.
- Deterministic rules around non-deterministic models — the validation layer in the SoundCloud recommender is a good example.
- Cost-aware LLM orchestration: staged, checkable steps over one giant prompt, as in MixLab.
- CI/CD and observability treated as part of the build, not an afterthought.
- Simple and maintainable beats clever, most of the time.
