AI Lead

About the Role

We're building an experimental system that uses LLM agents to automatically detect and patch vulnerabilities in software. Think updating packages, modifying Dockerfiles, recompiling binaries like Go with patched dependencies, etc. We don't yet know the perfect approach — this is research-heavy, and we're looking for someone who enjoys operating in that ambiguity.

This is not a traditional engineering role. We're looking for someone to live in prompting, workflows, chaining, validation, and iterating on LLM agents all day. You'll be designing agents that can reason about software dependencies, detect vulnerable components, propose fixes, and validate those fixes using other agents or tools.

Responsibilities

• Research and prototype LLM agent workflows for:
• Dependency graph analysis (e.g., Python, Go, Docker)
• Security patch recommendations and upgrades
• Automated recompilation and validation steps
• Design prompt chains and agent flows using LangChain or LangGraph
• Build validation loops using agents to assess agent-generated patches
• Run rapid experiments in notebooks or sandboxed environments — productionization is not your concern
• If the agent needs tools (e.g., to inspect Docker image layers), you should be able to:
• Write a basic utility yourself
• If complex, clearly specify what's needed for an Engineer to build it for you
  
  

• Strong understanding of software dependencies and vulnerability management
• Awareness of LLM evaluation techniques, hallucination mitigation strategies, and model tuning methods.
• Comfortable working in Jupyter, LangChain, LangGraph, or similar environments
• Familiarity with agent design patterns such as ReAct, Toolformer, LangGraph, or AutoGen multi-agent swarms.
• Fluent in experimenting and iterating — treats notebooks as a lab
• Enough scripting experience (Python preferred) to glue tools together
• Curious, independent, and not afraid to explore unconventional ideas
• Experience with context engineering — designing the inputs an LLM needs to reason accurately and efficiently (e.g., graph traversals, filtered SBOM diffs, CVE summaries).
• Skilled in tools engineering — building Python/Go utilities that agents can invoke (e.g., for image inspection, diff generation, or static analysis).
• Comfort architecting prompt+tool+validation loops with traceable state and measurable outputs.
  
  

• You don't need to be worried about DevOps, deployment, or fully productionizing - there will be help with that!
• This is not a frontend/backend/fullstack role. It is agent-centric R&D.