forgefield

ForgeField is a comprehensive construction management SaaS platform designed to streamline operations across drawing control, field execution, and administrative workflows. Built as an exercise in AI-native software development, the platform integrates artificial intelligence not as a bolt-on chatbot, but as core infrastructure for specific domain-heavy tasks. Key AI features include voice-to-structured daily logs that transcribe and parse foreman notes into drafts, COI compliance gates that validate subcontractor insurance certificates via PDF extraction, and drawing OCR with semantic spatial search using local embeddings. Additionally, it employs computer vision for auto-punch-pin re-projection across drawing revisions and includes an in-app AI assistant for guided onboarding. The platform offers over 35 modules organized into six pillars. Pillar 1 covers Field Ops & Safety, including quality inspections, pre-task plans with risk matrices, hot work permitting, and concrete maturity calculators. Pillar 2 addresses Contracts & Finance, handling change-order cascades, retainage escrow, certified payroll, and ERP synchronization. Pillar 3 focuses on BIM-VDC and data infrastructure, featuring real-time multi-cursor collaboration and a WebGPU IFC model viewer. Pillar 4 manages Preconstruction & QTO with tools for bid leveling and quantity takeoffs. Pillar 5 handles Commissioning, including closeout binder compilation and warranty lifecycle management. Pillar 6 provides Predictive Risk & Governance through executive dashboards, immutable audit ledgers, and cryptographic document watermarking. Technically, ForgeField is a single Next.js (App Router) and TypeScript monolith using PostgreSQL via Prisma 7. It employs Auth.js for authentication and Tailwind CSS for styling. The architecture supports offline-first field execution with a hand-rolled CRDT sync mechanism and uses PDFium-WASM or pdf.js for drawing rendering. All AI integrations are gated on API keys and fail soft, ensuring that missing credentials degrade features gracefully rather than causing application errors. The project explicitly documents its simplifications, such as using classical CV for punch-pin matching rather than learned models, and parsing IFC models instead of native Revit files.