Senior/Staff Applied AI Engineer — Conversational Agents

Revin · Agents Platform · Flatiron, New York City

$225,000–$325,000 base salary + meaningful early equity · Full-time

Four days per week in our Flatiron office, with one remote day and flexibility when life comes up. We are hiring at Senior and Staff levels. Final compensation depends on level, based on your demonstrated experience, scope of ownership, and impact.

About Revin

Revin builds AI-powered software for the businesses that keep our homes running: roofing, HVAC, plumbing, remodeling, and more. We help these businesses respond to customers, book appointments, and manage the customer experience across voice, SMS, and email.

Revin handles around half a million customer conversations a month. Businesses with hundreds of millions in annual revenue rely on our products for work that directly affects their customers and their bottom line.

Our customers include Vertex Service Partners, a national roofing and exterior-services platform with over $600M in revenue, and Del-Air, a $300M+ HVAC and plumbing business.

See the impact of our work: we’ve helped Del-Air book 11,000+ membership appointments, while ImproveIt uses Revin to set 34% of its booked appointments.

The role

Make our conversational AI agents more capable, responsive, and natural through rigorous experimentation. You’ll join our Agents Platform team at the intersection of software engineering and data science, using real customer conversations to understand where agents fall short and ship improvements that hold up in production.

You’ll own the full improvement cycle: identify a problem, form a hypothesis, design an experiment, implement a change, and validate its impact. Your work will span instruction following, tool use, task completion, latency, and conversational experience, with a substantial focus on voice.

You’ll improve agents through model selection, prompting, context, orchestration, and real-time voice behavior. We care most about your ability to learn from evidence and turn that learning into better customer experiences.

What you’ll do

Find the problems worth solving. Analyze production conversations, listen to calls, inspect traces, and work with customer-facing teams to identify failure patterns and prioritize improvements.

Build measurement you can trust. Define quality dimensions and representative evaluation datasets. Combine human judgment, automated graders, and production metrics, and validate that they measure what customers actually experience.

Develop and test hypotheses. Design reproducible experiments, compare approaches, and use offline evaluations and controlled production rollouts to understand what works, where it works, and what tradeoffs it introduces.

Improve the voice experience. Investigate turn-taking, interruptions, response latency, transcription errors, speech quality, and spoken naturalness. Use audio and system telemetry together to diagnose causes and validate changes.

Ship improvements into the agent system. Implement changes, monitor their effects in production, and take responsibility for reliability, cost, and regressions alongside quality gains.

Make improvement repeatable. Turn production failures and experimental findings into reusable datasets, evaluations, debugging tools, and release checks that help the team move faster.

At Staff level, you’ll also shape our evaluation and experimentation strategy across agent capabilities, identify the highest-impact opportunities, and raise the broader team’s ability to diagnose and improve agent behavior while remaining hands-on.

What you’ll bring

A demonstrated track record improving substantial AI agent systems deployed to external customers. You can explain the system’s scale and complexity, the problems you identified, your hypotheses and experiments, what you personally implemented, and how you established that it improved production outcomes.

Strong experimental and analytical judgment. You can define useful baselines, design informative comparisons, reason about noisy or confounded results, and learn from experiments that don’t support your hypothesis.