Julian Block
Building production AI
Built for systems that
can't afford to fail.
RAG, vision, and MCP — production AI for teams that need it to hold.
Previously: Goldman Sachs · Fidelity · World Series of Poker · ABF Freight · RoadOne
Philosophy
Great AI is a systems problem before a model problem.
A decade of shipping for banks, broadcasters, and freight networks teaches you the same lesson: durable AI products come from owning the entire path — ingestion, embeddings, retrieval quality, vision, orchestration, and failure handling. When those layers are designed together, the product feels effortless.
The weak versions fail in predictable ways: drift, latency spikes, brittle prompts, results that miss intent. The versions that survive Goldman Sachs-grade scrutiny feel composed because the architecture behind them is composed.
What I build
AI infrastructure engineered to survive production.
RAG pipelines with measured retrieval quality. Vision systems that classify and analyze at volume. MCP servers that let AI operate real infrastructure. The same architecture discipline I brought to finance, live broadcast, and freight — applied wherever the problem lives.
Retrieval
pgvector RAG pipelines — 10M+ embeddings, sub-second semantic search
Vision
YOLOv8 detection, segmentation & image analysis at production volume
Orchestration
MCP servers wiring LLMs into real systems, safely
Enterprise Platforms
Full-stack builds trusted by Goldman Sachs, Fidelity & WSOP
Featured system
Eden Earth — photo litter, clean it, see the ocean impact
