ML Engineer

About Willow

Willow is a voice-first productivity app that lets you type anywhere on your computer using your voice. People rely on Willow every day to write emails, respond in Slack, prompt AI, take notes, and move faster through their work.

But dictation is just our wedge.

We’re building the layer between people and computers: write with your voice today, operate software with intent tomorrow, and eventually enable workflows on your behalf with deep personalization.

We’ve proven adoption from individual power users to enterprises like Uber, Gusto, and Heidi Health. We’re backed by Y Combinator, Box Group, and founders from Instacart, Reddit, Shopify, and HubSpot.

We’re scaling quickly, and you’ll be joining early.

The Role

We’re looking for a Founding ML Research Engineer to build Willow’s personalization layer.

This role is for someone who believes the hardest unsolved problem in AI products is not raw model capability, but context, memory, user modeling, and deployment. Foundation models are getting better every month. The open question is how to build an AI system that understands a specific user in the flow of their work.

You’ll work on the systems that turn Willow from “great dictation” into a personalized communication and intent layer.

Today, a user can say something rough like: “Tell my boss I’ll come in at 6.”

Willow should understand the screen, the relationship, the app, the user’s tone, their past edits, their communication style, and produce the message the user would have written if they had taken the time to write it well.

That is the first problem.

Over time, the same systems that understand how a user communicates will increasingly help them operate software directly: navigating workflows, taking actions, automating repetitive tasks, and eventually acting on the user’s behalf in real time with deep personalization.

You’ll work across applied ML, product, inference, evals, and data systems to make Willow feel less like a tool and more like an extension of the user.

You’ll work directly with the CTO and eventually help build and lead a team across applied ML and research.

What You’ll Work On

You’ll help build systems for:

● Personal writing and behavior models that learn from edits, preferences, workflows, tone, formatting, and user actions

● Intent-to-action systems that turn rough user intent into high-quality execution

● Multimodal context understanding across voice, screenshots, app state, selected text, workflows, and historical behavior

● Retrieval and memory systems that understand what context matters in a given moment

● Realtime personalization systems that adapt to users continuously

● Evals for subjective quality: “does this feel like me?”, “is this what I meant?”, “would I trust this?”

● Agentic workflows that operate directly inside user workflows

● Feedback loops from edits, accepts, rejects, rewrites, and downstream user behavior

● The early foundations of systems that can proactively assist users in real time

This is not a pure research role and not a generic ML infra role. You should be excited to take ambiguous ideas, ship quickly, learn from real-world usage, and iterate aggressively.

What We’re Looking For

You are unusually excited by personalization, context, workflows, and user modeling.

You think deeply about questions like:

● How does an AI system learn what a user actually means?

● How do you build systems that feel genuinely personalized instead of generically helpful?

● How do you evaluate trust?

● How do you make realtime AI feel seamless enough to become part of daily workflows?

You probably have some combination of:

● Strong applied ML or research engineering ability

● Experience with LLMs, VLMs, agents, speech, personalization systems, retrieval, or multimodal systems

● Strong product intuition and taste

● Ability to ship production systems, not just research prototypes

● Obsession with latency, quality, and user experience

● A bias toward building systems that improve through real-world usage and feedback loops

Strong signals:

● You’ve built AI systems used by real users

● You’ve worked on personalization, memory, recommendation, agentic systems, speech, or multimodal AI

● You’ve shipped ambitious side projects, open-source work, research, or startups

● You care deeply about the future of human-computer interaction

We care much more about technical depth, product intuition, and speed than credentials or years of experience.

Why This Matters

The first major computer interface was the keyboard. The second was touch. The third is voice.

But the opportunity is larger than voice itself.

Voice matters because it captures intent naturally and with very low friction. By living directly inside user workflows, Willow gains something most AI systems do not have: continuous, real-world context.

We believe this becomes increasingly important as AI systems become more autonomous. Foundation models will become extremely good at long-horizon execution, but the highest-frequency interactions on a computer are different: fast, contextual, ambiguous, and deeply personal.

That requires understanding not just the task, but the user: their workflows, preferences, communication patterns, and intent in the moment.

Today, that means helping users write.

Tomorrow, it means helping users operate software.

Eventually, it means systems that can act on a user’s behalf in real time with deep personalization and trust.

Before You Apply

Try Willow. Use it in your real workflow. Dictate messages, emails, prompts, and notes. Think about what it would take for software to truly understand your intent.

We’re an intense, highly ambitious, in-person team in San Francisco. We optimize for learning, ownership, speed, and exceptional work. You’ll have significant responsibility, direct ownership, and the opportunity to shape the future direction of the company very early.

Don’t just submit an application. Reach out and show us how you think.