apps

Nash Paint Intelligent Analytics System

Predictive models for sales forecasting and operational planning, delivered through a web-based analytics interface.

Next.jsReactTypeScriptTailwind CSSFastAPIPythonSQLAlchemyPostgreSQLSQLiteChromaDBLangChainOpenAI GPT-4o miniOpenAI DALL-E 3OpenAI EmbeddingsPillowNumPyPydanticUvicorn

Deep dive

Case Study

Problem

Customers and store teams struggled to navigate Nash Paints’ wide product range (Decorative, Industrial, Automotive, Woodline, Accessories), choose the right system for each surface, match colors from photos, and find nearby branches with stock—relying on static catalogs and one-off consultations that didn’t scale.

Approach

Built a full-stack AI paint consultation platform: FastAPI backend with PostgreSQL/SQLite, ChromaDB + LangChain RAG for semantic product search, GPT-4o mini for conversational guidance, color extraction from uploaded images, virtual repaint previews, DALL-E 3 for visual concepts, and branch/stock data—all surfaced through a Next.js 15 + TypeScript UI with streaming chat, product cards, and dedicated chat, products, and repaint flows.

Results

Delivered an interactive Nash Paints AI Assistant where users can chat with a sales-oriented consultant, search products semantically, get color recommendations tied to real SKUs, preview paint on their own photos, and see rich product cards with prices and purchase links—turning catalog and expertise into actionable, in-app guidance.

Learnings

Grounding the model in the real product catalog (RAG, named SKUs, automatic product cards) and wiring recommendations into sales workflows (coverage guidance, primers, branch/stock context) mattered as much as raw model quality—closing the loop from AI output to purchase-ready next steps was what made the assistant useful in practice.