About

Esy (pronounced "Eh-see") runs agentic workflows that automate generation, quality-score outputs, and deliver approved artifacts at scale.

Templates define the pipeline. Agents execute it. Output is structured, auditable, and publishable.

The name comes from Essay Synthesis — "essay" in its original sense, from the French essayer: to attempt, to try. Synthesis is the pipeline that turns the attempt into a verified artifact.

The problem

Language models hallucinate citations. They reference papers that don't exist, fabricate quotes, and present invented data as fact. Image models produce anatomically wrong generations, mislabeled subjects, and artifacts that fail basic accuracy checks.

Most tools ship these errors directly to the user and call it done. No verification layer. No QA step. No audit trail.

The missing piece isn't better generation — it's a system that audits and catches these errors before anything gets published.

How it works

A template defines the workflow: what agents run, what verification steps execute, and what output format gets produced. Templates are predefined — users pick one, provide their sources or intent, and run it.

Agents handle the pipeline. Research, source gathering, citation verification, content structuring, and quality assurance all run in sequence. Each step feeds the next. No manual prompt chaining, no copy-paste between tools.

The output is a structured, publishable artifact with a full audit trail — not a chat transcript.

Principles

Built by

Zev Uhuru

Agentic Engineer — New York

I built ESY to evaluate whether agentic workflows can reliably produce artifacts that pass quality gates, across models, at scale, with humans in the loop for borderline outputs.

The first real test was clip.art, a platform I built to generate children's educational material at scale, used daily by my 4-year-old daughter. That became the production pipeline: 250–1,000 clipart, coloring pages, illustrations, worksheets, and infographics a day through provider routing, quality scoring, human-in-the-loop (HITL) review, and R2 delivery.

That work is ongoing. The same infrastructure is available to other engineers through my templates and the API.

GitHubYouTubeSubstackX

Questions? Email me at [email protected]