Project Foundry — AI Project Operating System
An AI-native platform for turning raw ideas into structured, actionable projects. Built with a TypeScript monorepo, React frontend, and an MCP server that integrates with Claude Code, Cursor, and any MCP-compatible client.
Challenge
Ideas fragment across notes, chat threads, and planning tools before they ever reach execution. Most planning tools don't integrate AI, and AI tools don't retain project context. The goal was to build the layer that closes this gap: a system where ideas are first-class entities that grow, get structured, and graduate into actionable projects — with AI acceleration at every stage.
Approach
End-to-end TypeScript monorepo with Drizzle ORM and PostgreSQL. React 19 + Vite frontend with shadcn/ui and TanStack React Query. OpenAPI spec as the single source of truth, with Orval generating React Query hooks and Zod schemas automatically. The AI layer includes a 5-stage Architect Mode, seven specialist assistant modes with SSE streaming, and an AI Operations Framework with 20+ governed operations. Context protection is enforced via a mandatory Context Protection Gateway (CPG) — all AI calls pass through classification-based field drop and PII detection before reaching any model. The MCP server exposes 70+ tools via JSON-RPC 2.0 with first-party OAuth 2.0/PKCE, making project context available to Claude Code, ChatGPT, and Cursor. Authentication supports Google OAuth, Apple Sign In, local email/password, and Microsoft Entra ID OIDC.
Outcome
Project Foundry is in active use as the operating system for all Darminsky projects — including the planning, execution, and documentation of the website you are currently visiting. The platform handles idea management, architect-guided planning, artifact generation, versioned documentation, enterprise auth, Stripe billing, GitHub integration, and real-time AI assistance. It demonstrates that a small team can ship a substantial, production-grade SaaS product with deliberate architecture and sustained execution.