AI-native media search, engineered for depth.

Pixel Detective transforms personal archives into a navigable intelligence layer: semantic search, real-time clustering, and curated latent-space exploration across tens of thousands of images.

Microservices + GPU UMAP Next.js 15 + Chakra UI Qdrant Vector DB
Read the Manifesto Explore the Platform
Abstract UMAP visualization

How it queries

Pixel Detective turns text and images into embeddings, searches Qdrant for semantic neighbors, and returns ranked results with metadata context.

Query flow diagram

Credibility, not hype

GPU-accelerated UMAP 10-300x speedups with CUDA and cuML.
Production-grade microservices Ingestion, ML inference, and vector search decoupled by design.
Interactive latent space Clustering, lasso selection, and curation workflows live.

Dive deeper

Follow the story: why it exists, how it works, and how the code evolves.

Manifesto preview

Manifesto

From promptcraft to production systems. The narrative behind the build.

 Pixel Detective architecture

Pixel Detective

Architecture, core services, and the platform that powers the search.

 Developer graph visualization

Dev Graph

Temporal intelligence layer for code, docs, and sprint evolution.

Planning DNA

Each milestone is grounded in PRDs, sprint plans, and architecture references. The system evolves with documented intent, not folklore.

Sprint 09-11 - Microservices refactor, GPU acceleration, latent space curation.
Architecture specs - Frontend and backend blueprints kept current.
Roadmap discipline - Features mapped to measurable performance goals.