POCAgentic AI
HumanAISentinel
MCP runtime security and guardrails POC
Personal project
Prototype of agent action and tool-security patterns.
A prototype guardrail layer for AI agents: policy evaluation, risk checks, human escalation, audit, and detection of unauthorized tool use.
Problem & business context
As AI agents start calling tools and APIs, organizations need runtime controls that stop risky actions and bring a human in when needed.
My role & contribution
Designed and built the prototype.
What I built or designed
- Policy evaluation and risk checks on agent actions
- Human escalation and audit trail
- Patterns for detecting unauthorized tool use
- MCP-style integration