LMS is the UK's leading provider of conveyancing services, processing 450,000 property transactions annually and connecting major banks, 4,000+ law firms, and property buyers through their digital platforms. I joined through Novus consultancy in January 2025 as Technical Architect, leading the design and implementation of core platform services while driving architectural standards across the organisation.

My primary focus was the NPTN (National Property Transaction Network), a digital ecosystem enabling communication between all conveyancing parties: material information providers, estate agents, law firms, and lenders. The platform is built on the PDTF (Property Data Trust Framework) schema, a UK standard for property transaction data interoperability - though I identified architectural risks with tightly coupling internal storage to an externally-controlled schema that undergoes frequent breaking changes.

The most significant technical contribution was a comprehensive authorisation service, evolved from patterns I'd developed at Spa Space. Property transactions involve complex multi-hierarchy relationships: individual users belong to branches, branches sit within districts, districts within brands, within organisations - and organisations are assigned as participants to transactions with role-based permissions. The existing implementation had authorisation logic scattered across eight services, each doing it differently, with every consumer having read/write access to both the organisation and transaction databases.

I designed and implemented a unified authorisation service supporting three models: role-based access control, fine-grained permissioning, and crucially, relational authorisation where a user's access to a transaction depends on their relationship to a branch, which has a relationship to a participant, which has a role in the transaction. The initial implementation took two to three weeks; the subsequent effort to maintain backward compatibility with the legacy approach took three months. Performance tuning was required when testing against production-scale data - initial queries took 10 seconds, which I reduced to 300 milliseconds through indexing and query optimisation.

I also developed the "Codebase Spider" - an automated documentation engine that pulls code from all 390 LMS repositories, analyses execution paths, and generates C4 diagrams and sequence diagrams using Structurizr. This has since evolved to include vectorisation of the entire codebase, enabling AI interrogation through an MCP server. Developers can now ask questions against the full codebase through RAG, or generate documentation on demand - a capability I'm actively refining using Claude.

Beyond technical delivery, I've contributed to team development through mentoring, code reviews, and requirement elaboration sessions - helping developers understand architectural patterns and advocating for proper refinement before sprint commitment.