Context
I led engineering delivery on an anonymised secure legal-tech SaaS platform designed for medical negligence document pagination and AI-assisted early case triage. The product started as a Vue application in AWS with Firebase support, then shifted mid-delivery into a tightly governed Azure estate as part of an organisation-wide platform migration.
The problem
Clinical review teams were manually reading large medical PDF bundles to decide whether claims were viable. That process was slow, inconsistent, and expensive. The product needed to automate pagination and extraction while maintaining strict confidentiality, role-based access, and auditable processing in a regulated legal workflow.
My role
I was the primary engineer delivering across front-end architecture, cloud integration, secure workflow implementation, and AI orchestration. I rebuilt the UI in React and Redux, integrated enterprise authentication, implemented document search, and collaborated with data scientists to operationalise model logic in Azure-based processing functions.
Approach
I treated delivery as both a technical and governance challenge: re-platform during active migration, redesign the frontend and identity path, and keep shipping value despite constrained tenant controls, tool churn, and changing stakeholder pressure. The delivered system used ADFS-authenticated React workflows, Azure Logic App orchestration, Python Azure Functions for AI/NLP processing, Cosmos DB for application state, and Elasticsearch for fast evidence retrieval.
Initial architecture (phase 1)
The first product version was a Vue-based application in AWS, supported by Firebase services. This validated product direction but did not complete full ML workflow integration before strategic platform changes redirected delivery.
Strategic platform shift
During delivery, core infrastructure moved to Azure under strict external governance. Environment controls, access constraints, and release friction were high, so the solution was re-architected to fit locked-down enterprise tenancy while still progressing MVP outcomes.
Delivery challenges beyond code
Major delivery friction included rigid environment models not designed for modern SaaS iteration, migration across Jira/Bitbucket and GitHub/Azure DevOps, repeated post-signoff scope expansion, stakeholder pressure mismatch, and leadership shifts that redirected effort toward V2 planning before V1 closure.
Key learnings
This work reinforced that AI products in regulated domains need equal strength in model capability, operational architecture, and delivery governance. Strong engineering alone is not enough when ownership and decision pathways are unstable.
