Skip to main content

Agentic risk-assessment prototype

SoftwareComplete

Developer and presenter. Generative AI agentic risk-assessment hackathon, London.

The agent assessing two vendors: Acme Analytics, rated low risk and approved automatically, and QuickMail SaaS, rated high risk and escalated to the security team.

A full-stack generative AI demo using Node.js, React and REST APIs, with the OpenAI API for LLM reasoning and retrieval-augmented generation, presented as a consulting-style investor pitch.

A rapid full-stack demo built for a generative AI hackathon on agentic risk assessment: an AI agent that assesses the risk of working with a vendor. Choose a vendor and the agent reads its profile (whether it processes personal data, its certifications, where its data is kept and its incident history), reasons over it with the OpenAI API, using retrieval-augmented generation to pull in the data it needs, and returns a risk level, a recommendation, a confidence and an explanation.

In the recording, a vendor with SOC 2 certification that keeps personal data in the EU is approved automatically, while one that processes personal data in the US without certifications is escalated to the security team.

I presented it as a consulting-style investor pitch to generative AI experts in London, and it later inspired my entry to the Open Agent 2026 hackathon.

Authorship

My role: developer and presenter.