AI Engineer, Quality (Evals)

Fieldguide•San Francisco, CA

1d•Onsite

About The Position

Fieldguide is establishing a new state of trust for global commerce and capital markets through automating and streamlining the work of assurance and audit practitioners specifically within cybersecurity, privacy, and financial audit. They build software for the people who enable trust between businesses. Fieldguide is based in San Francisco, CA, but operates as a remote-first company. They are backed by top investors including Growth Equity at Goldman Sachs Alternatives, Bessemer Venture Partners, 8VC, Floodgate, Y Combinator, DNX Ventures, Global Founders Capital, Justin Kan, Elad Gil, and more. The company values diversity and aims to build an inclusive, driven, humble, and supportive team. As an early-stage startup employee, you will contribute to building the future of business trust, making audit practitioners' lives easier by streamlining their work and improving work-life balance. Fieldguide is building AI agents for complex audit and advisory workflows in a rapidly transforming $100B+ market. Over 50 of the top 100 accounting and consulting firms utilize their services. As an AI Engineer, Quality, you will be responsible for the evaluation infrastructure that ensures the reliability of AI agents at an enterprise scale. This role focuses on establishing evaluations as a core engineering capability, including building a unified platform, automated pipelines, and production feedback loops to quickly evaluate new models across critical workflows. You will work at the intersection of ML engineering, observability, and quality assurance to meet rigorous customer standards. Fieldguide is hiring across all levels, with seniority calibrated during interviews. This specific role emphasizes in-person collaboration at their San Francisco, CA office.

Requirements

You are an engineer who believes that evaluations are foundational to building reliable AI systems, not a nice-to-have
Evaluation-first mindset: You understand that for an AI company, not being able to evaluate a new model quickly is unacceptable
AI-native instincts: You treat LLMs, agents, and automation as fundamental building blocks and parts of the craft of engineering
Data-driven rigor: You make decisions based on metrics and are obsessed with measuring what matters
Production-oriented: You understand that evaluations must work on real production behavior, not just offline datasets
Strong product judgment: You can decide what matters and why, without waiting for guidance, not just how to implement it
Bias to building: You move fast and build working systems rather than perfect specifications
Multiple years of experience shipping production software in complex, real-world systems
Experience with TypeScript, React, Python, and Postgres
Built and deployed LLM-powered features serving production traffic
Implemented evaluation frameworks for model outputs and agent behaviors
Designed observability or tracing infrastructure for AI/ML systems
Worked with vector databases, embedding models, and RAG architectures
Experience with evaluation platforms (LangSmith, Langfuse, or similar)
Comfort operating in ambiguity and taking responsibility for outcomes
Deep empathy for professional-grade, mission-critical software

Nice To Haves

Experience with audit and accounting workflows

Responsibilities

Design and build a unified evaluation platform that serves as the single source of truth for all of our agentic systems and audit workflows
Build observability systems that surface agent behavior, trace execution, and failure modes in production, and feedback loops that turn production failures into first-class evaluation cases
Own the evaluation infrastructure stack including integration with LangSmith and LangGraph
Translate customer problems into concrete agent behaviors and workflows
Integrate and orchestrate LLMs, tools, retrieval systems, and logic into cohesive, reliable agent experiences
Build automated pipelines that evaluate new models against all critical workflows within hours of release
Design evaluation harnesses for our most complex Agentic systems and workflows
Implement comparison frameworks that measure effectiveness, consistency, latency, and cost across model versions
Design guardrails and monitoring systems that catch quality regressions before they reach customers
Use AI as core leverage in how you design, build, test, and iterate
Prototype quickly to resolve uncertainty, then harden systems for enterprise-grade reliability
Build evaluations, feedback mechanisms, and guardrails so agents improve over time
Work with SMEs and ML Engineers to create evaluation datasets by curating production traces
Design prompts, retrieval pipelines, and agent orchestration systems that perform reliably at scale
Define and document evaluation standards, best practices, and processes for the engineering organization
Advocate for evaluation-driven development and make it easy for the team to write and run evals
Partner with product and ML engineers to integrate evaluation requirements into agent development from day one
Take full ownership of large product areas rather than executing on narrow tasks