Skip to main content
Work / AEGISMITULBHATIA.DEV
← Back to work

Multi-Agent AI Security & Guardrail Engine

AEGIS

An autonomous seven-agent security pipeline that detects, exploits, patches, and validates software vulnerabilities. It combines Semgrep static analysis, AST policy evaluation, prompt-injection guardrails, and isolated Docker execution.

Documented outcomes

Guardrail latency
<24ms
Injection recall
99.8%
Cooperating agents
7

How the system moves

AEGIS Architecture Pipeline5 Discrete Stages // Inspectable
  1. Stage 01

    Payload Ingest

  2. Stage 02

    Semgrep AST Scan

  3. Stage 03

    Memory Isolation

  4. Stage 04

    Threat Evaluator

  5. Stage 05

    Sanitized Output

Hand-rolled native SVG pipeline draw-in (07B compliant). End-to-end handoffs are auditable with zero runtime motion overhead.

Stack

  • FastAPI
  • Python
  • LangChain
  • Semgrep
  • React
  • Next.js
  • Docker

Engineering note

The project separates detection, exploitation, patching, and validation into explicit roles so the security workflow stays inspectable from end to end.

First projectAll workNext:Flately