I build AI execution governance systems for agentic and LLM-powered workflows.

My current work focuses on detecting and reducing execution waste in AI systems: coordination loops, repeated planning, false progress, low real work, evidence gaps, budget-aware execution, and audit-ready decision traces.

SBR — Synthetic Bureaucracy Recompiler

SBR is a deterministic analyzer for AI-agent workflow traces.

It validates native trace schema before scoring, rejects malformed or empty inputs, detects synthetic bureaucracy, and produces CI-grade reports with:

• Synthetic Bureaucracy Index
• workflow diagnosis
• failure modes
• production-risk badges
• gate decisions
• CI exit codes

SBR is not an LLM judge and not employee monitoring. It analyzes workflow structure and evidence-backed execution behavior.

Live project: syntheticbureaucracy.com

Public showcase: github.com/gulyetkin/synthetic-bureaucracy

PermissionLayer — Permission-aware AI Memory Gateway

PermissionLayer is a permission-scoped RAG gateway: policy runs before retrieval, field redaction runs before prompt construction, and every decision produces a hash-chained audit receipt.

Alpha reference implementation — not production-certified. Same data, same question, different user, different allowed context.

Repository: github.com/gulyetkin/permissionlayer

Neravo — AI Execution Governance Layer

Neravo is an execution governance layer for AI systems.

It is designed to govern AI execution before expensive or high-risk model calls happen, including routing, budget authorization, premium-worthiness checks, quality/failure detection, provider-aware execution, and audit proof.

Python TypeScript Next.js CI-grade validation AI agent workflows LLMOps AI governance trace analysis workflow risk scoring budget-aware model routing audit-oriented AI infrastructure

My work is focused on execution governance, not simple AI wrappers.

I build systems that analyze, govern, and verify how AI workflows execute.