Julian Block
Director of AI Engineering · Ayzenberg
Building AI systems that are deliberate, cinematic, and production-ready.
I design and ship end-to-end AI products: retrieval pipelines, orchestration layers, and polished frontends that make hard technical systems feel calm, premium, and trustworthy.
Philosophy
Production AI is a systems problem before it is a model problem.
Strong AI products come from owning the entire path: ingestion, embeddings, retrieval quality, orchestration, interface pacing, and failure handling. When those layers are designed together, the experience feels obvious.
The weak versions fail in predictable ways: drift, latency spikes, brittle prompts, or interfaces that expose uncertainty at the wrong moment. The durable versions feel composed because the architecture behind them is composed.
Articles
The writing lives on its own route now.
Research notes, build journals, and product articles now sit on a dedicated page instead of taking over the homepage flow.
Research note
Meta Brain2Query Feels Like Retrieval Discipline Disguised As A Research Demo
Build journal
Using SAM, Segmentation Masks, and FashionCLIP For Fashion Recognition At Silvr.ai
Product article
AI Products Need Better Pacing More Than They Need More Model Parameters
Infra article
MCP Is Quietly Becoming Infrastructure, Not A Novelty
