← Back to All Projects 
LangGraph 
GPT-4o 
FastAPI 
Next.js 
React 
TypeScript 
Supabase 
PostgreSQL LangChain 
Pydantic 
Docker 
Tailwind CSS shadcn/ui AI AgentComputer VisionPrompt Engineering ![Screenshot full size]()
AI Agent
Image Analysis Agent
LangGraph GPT-4o FastAPI Next.js React TypeScript Supabase PostgreSQL LangChain Pydantic Docker Tailwind CSS shadcn/ui AI AgentComputer VisionPrompt Engineering The Problem
- Product companies manage thousands of images that need consistent, structured tagging across categories like season, theme, colors, objects, mood, occasion, design elements, and product type.
- Manual tagging is slow and does not scale; existing tools lack confidence scores, validation, or structured taxonomy.
Approach
Built a full-stack AI agent with a LangGraph DAG pipeline (12 nodes): preprocessor resizes and validates the image, a GPT-4o vision call produces a structured description, then 8 parallel taggers run on the text using a controlled taxonomy.
A validator checks tags against the taxonomy, a confidence filter applies thresholds, and an aggregator produces the final TagRecord.
Results are persisted in PostgreSQL (Supabase) with a flattened search index for fast multi-category AND queries.
Frontend: single upload with real-time analysis, bulk upload with background processing and polling, search page with cascading filters.
Results
- Structured tags across 8 categories with per-tag confidence scores and needs-review flagging.
- Parallel tagger execution via LangGraph Send API reduces latency.
- Search with AND logic using PostgreSQL array containment and cascading filter options.
- Bulk upload with background processing and live polling.
- Full Docker deployment with one command.
Key result: Structured tags across 8 categories with confidence scores, parallel taggers, and AND search
Architecture & Flows
System Overview
Tools Used
LangGraphLangChainOpenAI GPT-4oPydanticPythonFastAPISupabasePostgreSQLNext.jsReactTypeScriptTailwind CSSshadcn/uiDocker