Shipping AI in ITSM • Cloud platforms • SaaS operations
Thirty years in IT Service Management, the last two spent shipping applied AI into production. I run technical operations, modernise legacy platforms, and build AI systems that solve operational problems analysts actually have, not the ones that demo well.
Most of my recent work has gone into Solvyr, an applied AI engine for ITSM that I designed and built from scratch. It's live, paying its way, and used daily — which informs almost everything I write here about what works and what doesn't.
Before all of this I co-founded Rightio, scaled it from a £5k investment to £15M turnover, and exited in 2023. That experience shaped how I think about technology: it has to earn its keep.
A suite of AI capabilities built into the Sostenuto ITSM platform, shipped over twelve months:
- Auto Triage — ticket classification at 99.8% accuracy in production
- Smart Resolution — RAG-powered resolution matching with Cohere Rerank 4 and HyDE query transformation
- Knowledge Creator — AI-assisted article generation from resolved tickets
- Comms Boost, Survey Sentiment, Call Intel, Vault, Shield, Hype, Studio — and more
Built on Azure OpenAI and Azure AI Foundry. Ranked first of 47 businesses in our portfolio AI Maturity Assessment.
The lesson from twelve months of shipping: the model is the easy bit. Data quality, governance, and analyst trust are where the real work lives.
Pragmatic over theoretical. Solutions that hold up in production, not just in demos.
Reliability before new capability. Uptime and recovery matter more than the next feature.
Data quality before automation. Clean data enables everything; dirty data breaks everything, faster with AI.
Explainability or it doesn't ship. If analysts can't see why the model decided what it decided, they won't trust it, and they're right not to.
Governance is a feature, not a tax. ISO 27001, audit trails, human-in-the-loop. The boring stuff is what makes the interesting stuff usable.
Technology serves people and process. Not the other way around. Ever.
| Domain | Stack |
|---|---|
| Applied AI | Azure OpenAI, Azure AI Foundry, Anthropic Claude, RAG architecture, vector search, Cohere Rerank, HyDE, prompt engineering, model evaluation, agentic workflows |
| Cloud & Infra | Azure (migrated from IBM Cloud), Kubernetes, Docker, Terraform, CI/CD, cloud architecture |
| Development | Python, TypeScript, React/Next.js, Node.js, REST APIs, PostgreSQL, SQL Server |
| ITSM | ITIL, incident/problem/change/knowledge management, service desk operations, SLA design |
| Ways of working | Spec-driven development with Cursor + Claude, rapid iteration, evaluation-led AI delivery |
Solvyr and most of my day-job work is proprietary, so this profile is the public-facing slice: personal projects, proof-of-concept work, and patterns I'm validating outside of production. Expect AI integration experiments, architectural sketches, and the occasional thing built to scratch a personal itch.
If you want to see how I actually think about applied AI in production, my writing on it lives at chriscarswell.co.uk.
If you're working on ITSM platforms, SaaS operations, cloud modernisation, or putting AI into production properly, happy to talk.
🌐 Website: chriscarswell.co.uk 🧭 Strategy: chriscarswell.co.uk/strategy
Views and content here are my own and do not represent any organisation's strategy, products, or intellectual property.