Textile ERP
Enterprise Resource Planning Platform
Video Demonstrations
Textile ERP & Enterprise RAG Topology
React 19 • Express 5 • Drizzle • PostgreSQL • Clerk • XGBoost
Excel → ERP Centralization
Centralizes orders, costs, margins and stock for 8 production employees — replacing the shared-Excel workflow with role-based Drizzle ORM/PostgreSQL syncing.
French AI Production Chat
Hybrid RAG assistant answering French-language questions about live production data — ChromaDB via a custom Python HTTP bridge, Ollama/Groq, with Postgres LISTEN/NOTIFY event-driven reindexing.
Decision Intelligence
XGBoost delay-risk scoring, pure-TypeScript OLS cost regression, and anomaly detection layered on live production data.
Screenshots & Interface
A real textile factory in Korba ran on shared Excel workbooks — 8 production employees juggling raw Italian/French workbooks for orders, costs, margins and stock, with no queryable source of truth.
Built a centralized, role-based ERP (Express 5 + React 19 + Drizzle/PostgreSQL, OpenAPI 3.1 → Orval codegen, Clerk auth + IDOR protection), then layered decision intelligence: XGBoost delay-risk scoring, OLS cost regression, and a hybrid RAG assistant (ChromaDB + Ollama/Groq) answering French questions over live production data.
Killed the shared-Excel workflow for the production team — raw Italian/French workbooks became structured, queryable knowledge with ML-assisted delay-risk and cost insights.
Technical Deep Dive
A contract-first monorepo (Express 5 + React 19 + Drizzle ORM/PostgreSQL, OpenAPI 3.1 → Orval codegen) built for a textile factory in Korba, replacing the shared-Excel workflow used by 8 production employees with a role-based web ERP centralizing orders, costs, margins and stock. Secured with Clerk authentication, RBAC and IDOR protection. On top of the data layer sits decision intelligence: XGBoost delay-risk scoring, pure-TypeScript OLS cost regression, and anomaly detection — plus a hybrid RAG assistant (ChromaDB via a custom Python HTTP bridge, Ollama/Groq) that answers French-language questions about live production data, with Postgres LISTEN/NOTIFY driving event-driven reindexing.