Kaishi Partners is partnering with an AI company to hire a Senior AI Product Engineer to build and scale production AI agents that help businesses manage marketing and growth.
This is a hands-on engineering role with ownership of agent architecture, runtime performance and backend integrations. You’ll build systems that work with live business data, recommend actions and execute them with appropriate human approval.
What you’ll do
- Design and build agent systems, including orchestration, tool execution, context, memory and failure recovery.
- Develop permissions, guardrails and approval workflows for agents interacting with business systems and budgets.
- Build evaluations using real customer workflows to measure accuracy, tool use and task completion.
- Improve reliability, latency and cost through model selection, retrieval, caching and concurrency.
- Build backend integrations across advertising, social media, CRM and analytics platforms.
- Shape engineering standards, lead technical design discussions and own delivery from concept through production.
What we’re looking for
- 5+ years of software engineering experience, with strong backend engineering and system design fundamentals.
- Experience shipping and operating LLM agents for real users.
- A strong understanding of agent orchestration, including state management, tool execution, retries and recovery.
- Practical experience building evaluations and using the results to improve agent performance.
- Evidence of reducing latency or cost in production, with clear measurements.
- The ability to structure ambiguous problems and take ownership of delivery.
Useful additional experience
- Multi-agent systems running in production.
- Large-scale data ingestion or integrations with business platforms.
- Experience across both established technology companies and early-stage startups.
- MCP, open-source contributions, or model fine-tuning and distillation.
Tech environment
Python, TypeScript, LangChain, LangSmith, MongoDB, ClickHouse and AWS. You don’t need experience with every tool, but you should be comfortable learning quickly.