AI Engineer

Katalyst is building the Revenue AI layer for global enterprise sales teams. We've built an AI agent core that runs chat, takes actions, and keeps Salesforce updated on its own for companies in construction, finance, aviation, and trucking. This role owns the engineering that makes that core accurate, fast, and reliable in production.

This is an AI infrastructure role, not a prompt-and-ship role. You'll work close to the models.

• Own the AI agent core that powers chat, actions, and automated Salesforce/CRM updates across every product surface
• Do harness engineering, ie build the systems that let agents take real actions inside Salesforce, Slack, and Gmail reliably, not just generate text
• Do loop engineering, ie how an agent checks its own output, retries, recovers from failure, and knows when a task is actually done
• Build and own our eval pipelines so we can measure accuracy and catch regressions before customers do
• Design and improve guardrails that keep agents safe and predictable on live enterprise data
• Work across multiple LLM providers, pick the right model for each job, and push accuracy up while keeping latency and cost down
• Build the context graph that gives agents real deal state to reason over instead of stale data
• Ship to production fast, on systems that enterprise revenue teams use the same week
• Work directly with the founders on architecture, priorities, and what to build next

What we're looking for

• Strong engineering fundamentals and experience building with LLMs in production
• You've built evals and know why they matter
• Comfortable with agent orchestration, retries, guardrails, and the messy parts of making AI reliable
• High agency. You find the problem and fix it without being handed a ticket
• Curious, fast, and want to work at the edge of what these models can do
• Bonus: experience with Salesforce or other CRM data, MCP, vector stores, or real-time data sync