Kos.ai
Founding AI Engineer
About this role
Kos is building AI agents to automate financial document processing for infrastructure and defense companies handling billions in capital. As Founding AI Engineer, you'll own the agent runtime, model orchestration, and evaluation systems that power this virtual finance employee.
What you'll do
- Own the agent runtime including computer use, tool orchestration, and memory loops for learning customer workflows
- Design ensemble routing layer to select optimal models per step while maintaining provider flexibility
- Build evaluation harness to catch regressions on recorded customer workflows before production
- Translate frontier agent research capabilities into reliable production systems
- Partner with applied team to convert practitioner errors into prompt, model, and routing improvements
- Set hiring bar and culture for the AI engineering team
What they're looking for
- Agentic systems and computer-use agents in production
- LLM orchestration and tool use design
- Evaluation design for open-ended tasks
- Fine-tuning, prompting, and retrieval tradeoffs
- Agent and LLM research implementation
- Python/systems programming
- Scoping complex problems end-to-end
- Production ML systems reliability
Benefits
- Medical, dental, vision insurance with minimal/no premiums
- 401(k)
- Visa sponsorship
Opens the application — the Jobs AI extension fills it for you. Set up autofill
Opens the official application on the employer’s site. No login required.
Kos.ai
Kos.ai builds AI-powered virtual finance employees that automate complex invoicing, contract work, and financial document processing for infrastructure and defense companies. The company is hiring software engineers, platform engineers, and AI engineers to develop core systems, agent runtimes, ERP integrations, and evaluation infrastructure.
- Website
- kos.ai
Likely interview questions
- Walk us through a production agentic system you've shipped—how did you handle agent reliability, failure modes, and the gap between research papers and what actually works reliably for customers?
- Describe your approach to designing an evaluation harness for open-ended tasks like invoice processing. How do you catch regressions on complex workflows before they hit production?