Skip to content

Hackathon, Nov 2025

Dependency Watcher

An LLM that queries live CVE data, so vulnerability checks happen in context.

View repository (opens in a new tab)Talk about this project

Recognised

At the JellyFaaS AI Hackathon, November 2025

Problem

Dependency vulnerability checks usually run out of band, in a separate scanner that a developer has to read. An LLM that can query live advisory data can answer the question in context.

Approach

  • A serverless Python microservice runs on JellyFaaS.
  • The LLM calls function tools that query the OSV API for live CVE data.
  • Function-calling workflows let the agent chain lookups to detect vulnerabilities without a human steering each step.

Architecture

  • Serverless Python microservice on JellyFaaS.
  • Agentic workflow using function calling against the OSV API.

Limitations

  • Built in a hackathon timeframe. It has not been benchmarked for detection accuracy against an established scanner.
  • Its vulnerability data comes from the OSV API.

Stack

  • Python
  • Tool use and function calling