Ex-Microsoft AI architect and principal engineer building production-grade AI systems across LLMs, agentic workflows, computer vision, probabilistic modelling, and cloud-native platforms.
I help organisations turn AI from experimentation into operational capability.
Recent work includes:
- Agentic AI systems and autonomous workflows
- Enterprise RAG and LLM evaluation platforms
- AI-enabled analytics systems delivering multi-million-pound savings
- Computer vision systems using YOLO and multimodal pipelines
- Decision intelligence platforms for high-uncertainty environments
- Model optimisation, distillation, and scalable inference architectures
Designing production-grade AI systems using:
- LLMs
- RAG
- AI agents
- tool orchestration
- evaluation pipelines
- cloud-native inference architectures
Helping organisations reduce dependency on expensive frontier-model inference through:
- model distillation
- workflow optimisation
- hybrid architectures
- targeted local inference
Building systems that convert incomplete information and expert judgement into quantified, defensible decision outputs.
Example: 👉 https://www.darach.ai/risklens/
Decision intelligence for high-stakes uncertainty.
RiskLens converts expert judgement into quantified probability ranges, scenario analysis, and board-ready decision outputs using:
- Bayesian modelling
- probabilistic simulation
- expert elicitation
- uncertainty modelling
Designed for:
- infrastructure
- investment decisions
- strategic planning
- operational risk
- rare-event analysis
🔗 https://www.darach.ai/risklens/
A repository exploring practical strategies for:
- reducing LLM inference costs
- targeted model distillation
- hybrid model architectures
- evaluation workflows
- post-frontier AI delivery strategies
🔗 https://github.com/garyshort/llm_distil
Designed and delivered Azure-based multimodal AI systems that:
- converted damage imagery into supplier-specific estimates
- automated insurance documentation generation
- combined computer vision with LLM workflows
- integrated RAG and structured document generation
Technologies: Python • Azure AI • YOLO • MLflow • Kubernetes • Azure OpenAI
Architected AI and analytics systems for infrastructure and asset-management decision making.
Delivered:
- probabilistic risk modelling
- AI-powered inspection analysis
- synthetic training data generation
- operational optimisation systems
Estimated annual impact: ~£15M–£20M savings
Python • PyTorch • TensorFlow • scikit-learn • OpenCV • YOLO • MLflow • RAG • Agentic AI • Azure OpenAI • Llama • Phi • GPT
Azure • Kubernetes • Docker • Azure ML • Databricks • Synapse • Data Factory • Event Hub • CI/CD
Python • C# • TypeScript • Go • Rust • Scala • Spark • APIs • distributed systems
Previously:
-
Cloud Solution Architect at Microsoft
-
Microsoft C# MVP (6 years)
-
Principal AI/Data Science consultant across:
- insurance
- infrastructure
- mobility
- humanitarian analytics
- enterprise AI transformation
Fully hands-on.
I build:
- production systems
- prototypes
- architecture
- evaluation frameworks
- cloud platforms
- AI workflows
- technical strategy
Current tooling:
- Cursor
- Claude Code
- modern AI-native engineering workflows
- Agentic AI
- LLM systems engineering
- AI architecture
- inference optimisation
- probabilistic modelling
- AI governance
- cloud-native AI platforms
- multimodal systems
- decision intelligence
- synthetic data generation