Founding Research Engineer

About the company

Our client is a seed-stage applied AI lab in San Francisco, founded in 2025, with around ten people and $8M raised. It is building what it calls the zero-touch enterprise: context-aware agents that work inside heavily customised enterprise resource planning systems so that finance and operations workflows run on their own. The agents are already deployed at Fortune 500 companies across consumer goods, distribution, high tech and oil and gas, and the team comes from DeepMind, Meta AI, Amazon AGI Labs and the enterprise software world.

The role

This is a zero-to-one job on both the product and the infrastructure under it: the secure connector and runtime that understand a complex enterprise system, the developer tooling around them, and the analysis and planning engine that transforms legacy code. You would design the agentic layer itself — the self-checking, reflection and verification passes that run before a result ever reaches a human or a downstream system — and lead the work on hard guarantees that an agent never takes a harmful action. Much of it happens on live customer pilots, on systems that are out of distribution for today's models, so you would work with customers directly. The team is looking for strong early-career engineers rather than PhD graduates.

What you'll do

• Own a zero-to-one product and its infrastructure: the secure connector and runtime that read a complex enterprise system, the developer tooling, and the engine that analyses and transforms legacy code
• Design the agentic layer: self-checking, reflection and verification passes that run before results reach humans or downstream systems
• Lead the work on hard guarantees that agents never take a harmful action
• Build agentic workflows and retrieval-augmented generation, low-latency inference paths with token-level observability, and prompt pipelines, retrieval and light fine-tuning in production
• Improve agent performance on end-to-end Fortune 500 processes, sometimes inside live customer pilots, working with customers directly

Nice to have

• Software engineering at a high-growth AI startup
• Published machine learning research papers
• An ownership mentality, crisp communication, and comfort deciding on 80% of the information

Tech stack

Agentic workflows and retrieval-augmented generation, with LangGraph a plus; low-latency inference paths with token-level observability; prompt pipelines, retrieval and light fine-tuning running in production; and the secure connector and runtime that read and act inside a customer's enterprise resource planning system.

Compensation & logistics

• $175k–$300k base, plus 0.5–1.2% equity
• On-site in San Francisco five days a week, in an office next to the Caltrain station
• The client can support a visa transfer, such as OPT or an H-1B transfer, and can also file a new visa sponsorship for this role
• One to two hires on this role
• Interview process: a 30-minute initial screen on background, interest and culture fit, a one-hour technical deep dive and problem-solving interview, then a three-hour on-site with the founders
• Full-time