Senior .NET / enterprise systems developer focused on legacy modernisation, practical GenAI adoption, agentic workflows, and secure AI-assisted delivery.

My current portfolio explores how LLMs can be used safely and usefully in corporate software delivery: understanding existing systems, recovering requirements, defining interfaces, generating structured artefacts, integrating deterministic business rules, and building auditable AI-assisted workflows rather than black-box demos.

It also includes deliberately small legacy-enterprise refreshers, including WPF and COM Interop, because useful delivery often means bridging modern tooling into systems that are already running the business.

• maf-doc-processor - .NET 10 document-processing API built on Microsoft Agent Framework, with multimodal classification and extraction, typed workflow routing, composite multi-document capture, deterministic validation and policy, bounded repair and concurrency, usage/correlation telemetry, generated OpenAPI, and an accessible correction UI.
• interop-demo - classic .NET Framework COM Interop demo with a VBScript client, registration scripts, and explicit 32-bit/64-bit automation notes.
• sample-agent - portable email-to-report agent pattern that turns unstructured business requests into governed report outputs with clarification, permission, generation, response-drafting, and audit steps.
• loop-design-build - provider-agnostic harness for bounded human-in-the-loop agentic development, using Git-tracked markdown state to keep progress auditable and stop on ambiguity.
• Sample-NYCTraffic-Refresh - legacy-refresh case study showing how the loop-design-build harness supported system discovery and delivered a Node/React/SQLite operational data exploration tool in narrow, reviewable slices.
• office-vba - Excel and Access modernisation lab showing source-controlled VBA, object inventories, characterisation tests, generated SDLC artefacts, packaging scripts, and .NET migration spikes.

The earlier csharp-semantic-document-processor remains public as a frozen, deprecated Semantic Kernel predecessor. Together the repositories show the rewrite from Semantic Kernel to Microsoft Agent Framework and the subsequent extension of the system through an auditable agentic-coding delivery process.

• How to separate LLM reasoning from deterministic software responsibilities.
• How LLMs can support legacy system discovery, interface extraction and characterisation before modernisation work begins.
• How old-enterprise integration boundaries such as COM automation can be isolated, tested and documented.
• How to put typed contracts, policy checks, tests, and audit trails around AI-assisted workflows.
• How agentic development can be made observable and interruptible rather than treated as autonomous magic.
• How enterprise teams can evaluate practical AI adoption through small working slices, not slideware.
• How product-facing AI concerns such as prompt injection, data exfiltration, access control, auditability, and provider portability affect architecture.

• Secure AI-assisted SDLC workflows.
• LLM-assisted legacy discovery, reverse engineering and modernisation planning.
• Agent orchestration patterns that remain portable across tooling vendors.
• C#/.NET integration with LLM and multimodal model APIs.
• Governance boundaries between internal engineering AI use and AI embedded in delivered products.
• Turning vague business requests into structured, testable, maintainable systems.