SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Finch is building infrastructure to make legal services radically more accessible, starting with personal injury law. The company has grown 10x in just over a year, raised a $20M Series A, and is backed by Sequoia, Redpoint, and founders of companies like DoorDash and Ironclad. They're automating the administrative work in pre-litigation, from intake and claim opening to medical records, lien management, and demands.
As a Member of Technical Staff in Applied AI, you'll own critical features end-to-end in a full-stack role where product evolves quickly and engineers have real ownership. Your responsibilities include:
- Build production-grade voice, browser, document, and workflow agents that move personal injury cases forward
- Design long-horizon agent systems with context, tools, memory, permissions, structured outputs, and human-in-the-loop controls
- Own the eval flywheel: turn real cases and operator feedback into benchmarks that continuously improve agent quality
- Optimize performance through model selection and routing, context engineering, prompting, tool design, and agent observability
- Make agents reliable across calls, documents, communications, third-party portals, partial failures, and changing case information
- Evaluate advances in agentic AI and rapidly ship improvements to quality, speed, or cost
You'll work directly alongside product, ops, and design teams in a codebase where your decisions matter from day one. The role is 4 days/week in the NYC office.
REQUIREMENTS:
Must-haves:
- 3+ years building reliable, secure software in production, with ability to own systems across the stack
- Experience shipping AI-powered products or agents beyond the prototype stage
- Experience designing practical evals and using them to understand agent behavior, compare approaches, and improve task-completion quality
- Proficiency in Python and comfort working with React, LLM APIs, and agent frameworks
- Strong instincts for agent design, including context construction, model selection, tool boundaries, structured outputs, failure recovery, and human handoffs
- Track record of turning ambiguous, real-world workflows into simple, dependable systems
- Experimental mindset: move quickly, measure what matters, care more about whether an agent completes the job correctly than whether the approach is clever
Nice-to-haves:
- Experience with Pydantic AI or another production agent framework
- Experience building voice agents, browser agents, or long-running workflow agents
- Familiarity with Django, React, Temporal, Playwright, or AWS
- Experience building evaluation systems, agent observability, model routing, or human-in-the-loop review tools
- Familiarity with legal tech, healthcare workflows, document processing, or other operationally complex domains
