Benjamin Fielder

AIDE: AI-Driven Engineering Platform · C#, ASP.NET Core (.NET 10), EF Core, PostgreSQL, React 19, TypeScript, SignalR, Kubernetes, GitHub Actions

A full-stack platform that turns an idea into released software through a governed six-phase lifecycle: refinement, specification, planning, execution, review and release. AI agents produce versioned artifacts and people approve them at gates. Rebuilt on .NET from an earlier TypeScript/NestJS prototype, using my own react-dotnet-local-kit.

• Layered ASP.NET Core architecture: business rules (lifecycle state machine, approval gates, artifact versioning and rollback, review scoring) live in a pure Domain project, with thin API endpoints and EF Core code-first migrations on PostgreSQL.
• Five AI agents behind a single LLM provider abstraction (Anthropic, OpenAI, Ollama, plus a deterministic fake for tests), with per-agent model selection, token budgets and cost tracking.
• Background workers claim jobs from PostgreSQL with FOR UPDATE SKIP LOCKED, and a transactional outbox writes each state change and its follow-up jobs atomically, with no separate message broker.
• Git integration generates code step by step, opens pull requests, runs review cycles, then merges and tags releases. SignalR pushes live updates; notifications go out by email, Slack, Teams and HMAC-signed webhooks; an aide CLI ships as a .NET global tool.
• React + Vite frontend with TanStack Router and Query, calling the API only through a TypeScript client generated from the OpenAPI contract.
• Tested with xUnit (unit, and integration via Testcontainers), Vitest with MSW, and Playwright end-to-end with axe accessibility checks. GitHub Actions CI enforces warnings-as-errors and fails on OpenAPI contract or EF migration drift, then deploys to local Kubernetes with Kustomize, a pre-rollout migration Job and smoke tests.

claude-kits: Claude Code plugin marketplace · Claude Code, Plugins, Agent skills, Custom agents

A public, installable marketplace of Claude Code plugins, each a "kit" of skills and agents tuned to one tech stack, so AI-generated code follows the same conventions every time. Three spec-driven kits so far, with 21 skills and 30 subagents between them.

• react-dotnet-local-kit: React, Vite, ASP.NET Core, EF Core, PostgreSQL, GitHub Actions and Docker Desktop Kubernetes. Used to build AIDE.
• react-dotnet-k8s-kit: React, .NET, GitLab, Kubernetes, Argo CD and Jira, mirroring my professional stack.
• nextjs-supabase-kit: Next.js, Supabase and Vercel. Used to build Impulse.
• Each kit pairs an orchestrator skill that builds from a spec with stack skills (ASP.NET Core endpoints, EF Core data, OpenAPI contracts, CI, Kubernetes deployment) and role-based subagents for each stage of the lifecycle: scaffolding, engineering, QA, PR review and spec verification.
• Run like a released product: versioned plugin manifests, validation with claude plugin validate before every push, and update channels for installed users.

Impulse: Training-programme planning SaaS · Next.js, TypeScript, Tailwind CSS, Supabase, PostgreSQL, Vercel

A web app for coaches and athletes to plan periodised gym-and-pool training programmes, built from my own experience as a national-level swimmer.

• Periodised programmes (phases, weeks, sessions) modelled in a relational PostgreSQL schema with row-level security and Supabase auth.
• Full product loop: authentication, a programme builder UI and server-side data access, deployed on Vercel.
• Built with my own nextjs-supabase-kit and Claude Code, reviewing and owning all generated code.

Portfolio, web apps & games · Next.js, React, three.js, p5.js, Unity/C#

Built and deployed benfielder.com, Sketch Vault and a WebGL game portal on Vercel with custom domains and PR-based workflows. Developed Unity/C# mobile games under Space Worm Studio, taking Pyramid Numbers through to Google Play submission.