I'm Will Leece,
a full-stack .NET developer based in the UK, currently building at Central Technology.
About
I build .NET web apps, from real-time SignalR wallboards to multi-tenant client portals, and I specialise in AI-output verification: building independent QA and checks around LLM-generated content rather than trusting model output directly.
Experience
- Mar 2025 – Present
.NET Developer at Central Technology
Remote · Chesterfield-based company
- Build production AI features with Microsoft Agent Framework across a range of use cases, from chat interfaces to recommendation generators.
- Set up AI-assisted development workflows in Codex and Claude, using optimised instruction files, execution plans and git worktrees. Running two to three workstreams in parallel at least doubles my throughput.
- Built an AI benchmarking suite, packaged as a Codex skill, that compares any model available in Codex on a common task, and tests instruction files per model per task to find the best model and instruction file for each job.
- Built SignalR wallboards and a reusable framework that gets a new branch its boards in about 20 minutes.
- Build features for a multi-tenant client portal used by hundreds of client organisations and thousands of users, built in ASP.NET Core, Blazor and EF Core, where customers view their own data and submit requests.
- Optimised many slow reports and queries, some from minutes to seconds, using parameterised OPENQUERY, subquery filtering and EF Core refactoring.
- Oct 2022 – Mar 2025
Junior .NET Developer at Central Technology
Chesterfield, UK · Hybrid
- Built bespoke features across internal and customer-facing CRUD applications using ASP.NET Core, C#, and EF Core.
- Authored bespoke reports for recurring operational issues and data forensics that turn manual investigation into repeatable tooling.
Projects
Meducate API
A REST API that ingests medical data from MedlinePlus and PubMed, uses an LLM to classify and structure it into standardised medical categories, and serves it through a versioned interface refreshed daily. Nightly quality-control jobs check that AI-generated summaries haven't drifted from their source material and flag any that have for re-processing. Diagnosable topics are also cross-referenced against the WHO ICD-11 API and tagged with a real diagnostic code where a confident match exists.
Wayfarer
A natural-language query tool that plots results on a map, grounded in Wikidata rather than guessed by an LLM. An agent explores Wikidata's schema live and builds and tests a SPARQL query, then a verification gate checks the result against expected type, row count, and a second independent pass before anything reaches the map. Anything that can't be confirmed gets flagged in the UI instead of being blended in with the verified results.