Hugo Frauches
ENPTLet’s talk

Selected work

Three systems in production.

Case studies of AI products I conceived, helped build and led. No client names, and no private architecture.

01 · Security · In production · May 2026

AI external attack surface intelligence

ISA Horizon

Problem

Teams need to see what is exposed, and which credentials are already in circulation, before someone else uses them.

Outcome

In production since May 2026. The control described publicly is human validation before a finding is released.

Approach

An agent reasons about external exposure the way an attacker would: it maps what is exposed and which leaked credentials are tied to it. A person validates each finding before it reaches the customer.

My contribution

  • Led conception
  • Contributed code
  • Worked on architecture
  • Led the development team

Visit product website ↗ (ISA Horizon)

02 · AI · In production · May 2026

AI vulnerability operations copilot

ISA Insight

Problem

Findings arrive from more than one scanner. The hard part is deciding what to fix first, and how.

Outcome

In production since May 2026.

Approach

A copilot brings those findings into one risk-prioritized remediation plan and lets the team ask questions about them in natural language.

My contribution

  • Led conception
  • Contributed code
  • Worked on architecture
  • Led the development team

Visit product website ↗ (ISA Insight)

03 · Product · In production · June 2025

Agentic AI SOC platform

ISA Cyber

Problem

Alert volume does not pause. Triage, enrichment, investigation and correlation have to continue through the night, without giving up the judgment that stays with a person.

Outcome

In production since June 2025. The published result is qualitative: faster detection and response, resolution SLAs met in nearly all cases, and a lower cost of triage without adding people.

Approach

Agents triage, investigate and correlate alerts around the clock, including signals from the security tools a team already runs. They draw on runbooks, past incidents, advisories and threat intelligence, and each customer’s data stays separate. The work is judged on precision, speed, response commitments and coverage of real incidents. Safeguards, limits on what an agent can reach, and a person’s approval for high-impact actions are part of the design.

My contribution

  • Led conception
  • Contributed code
  • Worked on architecture
  • Led the development team

Visit product website ↗ (ISA Cyber)

Product names belong to their owner. No client data, and no private architecture, is published here.