Pixel Sprites
Multimodal Memory & Agentic Content System
Stateful LangGraph workflows classify tasks, understand media, plan content, generate keywords, and validate outputs before returning them.
Overview
A multimodal AI application with specialized agents for understanding and creating content from images, video, and audio.
Highlights
- Used structured Pydantic outputs, conditional routing, and validation retries to keep generated content aligned with requirements.
- Configured OpenRouter model tiers for text, vision, and reasoning; added LangSmith tracing for workflow inspection.
- Built media indexing and memory with SHA-256 identifiers, metadata extraction, Supabase persistence, and InsightFace embeddings.
Tools and methods