Senior Applied AI/ML Scientist - Compass

Faire•San Francisco, CA

20h•$196,000 - $269,500•Hybrid

About The Position

Faire is building the future of wholesale, connecting independent retailers with the brands that will define their stores. At the heart of this mission is Compass — Faire’s user facing AI bet within the Discovery Pillar — building an always-present, context-aware retailer assistant along the enagement journey. Compass helps retailers make smarter buying decisions by combining Faire’s rich proprietary data with agentic AI and web search, and is increasingly able to take action on retailers’ behalf. As a Senior Applied AI/ML Scientist on the Compass team, you will be the science and technical lead for this product — driving agent quality through data, evaluation, and modeling, while shipping product features end-to-end with high velocity. This is a deeply hands-on individual contributor role: no direct reports, keyboard first. You will set the data-grounded direction for how the assistant works, while also being a full-stack (AI, ML, backend) builder who turns ideas into shipped product fast. You will work at the frontier of agentic AI, blending applied science rigor (eval-driven development, experimentation, data strategy) with cross-stack engineering range to build the retailer assistant of the future. This is a rare opportunity to shape a product from near-zero — where your judgment, speed, and instincts will define the outcomes.

Requirements

5+ years of industry experience building and shipping production ML/AI systems with measurable business impact — including hands-on ownership of the applied-science side (data, evaluation, modeling, quality), not just system plumbing.
Has shipped agentic / LLM-powered features in a core production product — with a deep, opinionated grasp of agent design tradeoffs: eval strategy, latency/cost/quality tension, tool-calling vs. context preload, guardrails, and failure containment.
Strong applied ML / data science foundation — reasons from data, designs experiments and evals, and has turned proprietary or structured data into product capability.
Track record of shipping fast across multiple stacks (backend, data, and ideally frontend) with quality — not a single-layer specialist; demonstrates cross-stack range.
AI-native in practice: uses AI coding tools and agent workflows as a force multiplier in day-to-day work.
Architectural maturity — can explain design choices that work simply today but won’t need to be thrown away when requirements grow.
Operates with high autonomy and resourcefulness, with good judgment about when to escalate and when to just solve it.
Fluent enough in engineering to make sound architecture calls.

Nice To Haves

E-commerce, marketplace, or two-sided platform context — understanding of both sides of the retailer/brand dynamic.
Experience evolving a read-only assistant into one that takes actions safely — confirm-first patterns, guardrails, and failure containment.
Hands-on experience with the OpenAI Agents SDK or similar agentic frameworks in production.
Familiarity with preload-over-RAG context strategies, Snowflake-backed grounding, or hybrid approaches.
Prior 0→1 / early-stage product experience — has built something meaningful from scratch.
Recommendation, retrieval, or personalization modeling background.
Public writing, open-source contributions, or talks that show structured thinking about agentic / applied-AI systems.

Responsibilities

Own the science and technical north star for Compass’s agentic products — the retailer assistant today and whatever comes next: how to leverage Faire’s proprietary data, agent + tool + context strategy (preload vs. tool-calling vs. hybrid), and how to measure and raise agent quality as systems gain the ability to act.
Ship retailer-assistant features end-to-end — across the FLARE Python app, data plumbing, tool wrappers, and the frontend surfaces where the assistant appears, using AI-native workflows to multiply your output.
Translate ambiguous product bets into sequenced, de-risked tactical plans — what to build now, what to defer, and which bets carry the highest impact × probability-of-success.
Set and raise the bar for eval- and experiment-driven development — define how the team knows an agent is good, including offline eval suites, LLM-as-judge metrics, and quality criteria per surface and retailer journey.
Make pragmatic engineering choices: simple enough to ship now, designed to evolve — not over-engineered for imagined future scale, but not throwaway either.
Partner closely with engineers on architecture and serving tradeoffs, and act as the science/technical interface to adjacent teams (Search, Personalization, Platform/FLARE).
Raise the team’s collective judgment through prototypes, analyses, design reviews, and pairing.