Zyberpol — agentic SOC platform

Co-founded VEZRAN to build agentic AI for security operations. Leading AI/ML architecture for Zyberpol, the flagship product.

What Zyberpol does

Zyberpol sits on top of an existing security stack — CrowdStrike, Okta, Splunk, AWS — and runs four working agents under a configurable autonomy model:

• Correlation — stitches signals across data sources into incident hypotheses
• Investigation — pulls supporting evidence and rules out false positives
• Triage — ranks and recommends actions
• Remediation — executes approved actions and writes the evidence trail

By default the system recommends and humans authorize. Autonomy is configurable per action type, per environment, per analyst. The differentiating bet: every action ships with a signed, timestamped, audit-ready evidence package — the “PROVE” layer that downstream insurance underwriters and audit teams require.

My scope

• Multi-agent orchestration on frontier LLMs (Claude, GPT-4o, Gemini) with deterministic guardrails
• RAG over security context — alert history, runbooks, threat intel, environment graph
• Evaluation framework for measuring agent correctness when ground truth is incomplete (see related essay)
• Production infrastructure: latency, cost, fallback paths for tier-1 triage delegation

Why now

95% of 2025 intrusions used automation. Human-paced defense is mathematically losing. Frontier reasoning is finally reliable enough to delegate tier-1 triage, and cyber insurance underwriters now require provable controls — 40%+ of 2024 claims were denied for unprovable response.

Pre-seed, raising $4M, hiring across AI/ML and security engineering. vezran.com