---
title: 'Founding Machine Learning Engineer at Composite'
canonical: 'https://feeny.ai/job/founding-machine-learning-engineer-composite-san-francisco-sj1kfbk2jwbb'
type: 'job'
last_seen: '2026-09-08'
---

# Founding Machine Learning Engineer at Composite

- **Company:** Composite
- **Location:** San Francisco, CA
- **Compensation:** $150k–$220k
- **Employment:** full-time
- **Work type:** onsite
- **Posted:** 2025-10-22
- **Last confirmed live:** 2026-09-08
- **Apply:** https://jobs.ashbyhq.com/composite/03e445d5-f7c9-41f3-8994-ed136dc1e64d

## Job description

## ABOUT THE ROLE

We're looking for founding Machine Learning Engineers (MLEs) to own and improve our core action models end-to-end - the intelligence that powers Composite's proactive automation platform.

You'll work at the intersection of LLM inference, browser understanding, and low-latency systems, shipping models that need to feel instant while reasoning over complex page state and user context.

Unlike hosted browser solutions that introduce latency and auth barriers, or consumer-focused "AI browsers," we run AI directly through professionals' existing browsers via a Chrome extension, creating instant response times with zero migration or IT friction. This architecture creates unique ML challenges.

This is a high-ownership role on our small, exceptional team where your work ships directly to users and has the potential to tangibly improve the work lives of hundreds of millions of people.

## ABOUT COMPOSITE

College-educated professionals spend 85% of their day as digital factory workers in Chrome, clicking through repetitive browser tasks.

Composite is building the proactive layer for productivity so professionals around the world can focus on meaningful, high-leverage work. We're training action prediction models that run in real time, anticipating what you'll do next based on page context and prior interactions.

We've raised $5.6M in seed funding led by Nat Friedman and Daniel Gross, with participation from Menlo Ventures, Anthropic's Anthology Fund, SVAngel, and other incredible investors.

## WHAT YOU'LL WORK ON

- Improve the accuracy and latency of our core models across diverse web applications to predict users' intended next actions and execute them faster than manual input
- Design and optimize LLM inference pipelines, including token caching strategies, streaming architectures, and network-level optimizations between client and server
- Build evaluation frameworks and data pipelines to measure and improve model quality at scale
- Experiment with retrieval-augmented approaches using vector databases for contextual memory
- Develop synthetic data generation pipelines for browser interaction training data
- Work with DOM states, accessibility trees, and user interaction data to improve browser understanding
- Ship features end-to-end that go directly to users — this is not a research-only role

## WHAT WE'RE LOOKING FOR

## ML & SYSTEMS

- Strong ML fundamentals with hands-on experience training and deploying models in production
- Obsessive about latency — experience optimizing inference pipelines to feel instant to end users
- Deep care about data quality, with the instinct to build tooling that ensures it
- Experience with LLMs, transformer architectures, or sequence prediction problems
- Comfortable working across the stack — our system spans a Chrome extension, Electron app, Cloudflare Workers edge proxy, and inference providers

## CORE QUALITIES

- Character: You're someone we'd want to work closely with for the next ten years. You approach challenges with curiosity rather than ego. You're a team player, a great communicator, and aren't afraid to be wrong.
- Work Ethic: You're energized by hard problems and comfortable working intensely toward ambitious goals.
- Raw Intelligence: You can quickly understand complex systems and solve novel, ambiguous problems with self-guidance.

## BONUS

- Experience with browser automation, Chrome extensions, or web scraping at scale
- Familiarity with accessibility tree / DOM parsing for page understanding
- Background in RL or online learning from user interaction data
- Experience with vector databases (e.g., Turbopuffer, Pinecone) and hybrid search
- Full-stack development experience (TypeScript, Node.js, React)

## OUR VALUES

- Disagree and commit: Respectfully challenge decisions you disagree with, even when it's uncomfortable. Don't censor yourself or your ideas. Once a decision is determined, everyone commits wholly.
- Clear and consistent standards: Decisions are made based on a shared framework that applies for everyone. We don't leave room for "rules for you, not for me" or any perceived hypocrisy.
- Over-communicate: Nothing slows down a company more than confusion, mis-, or under-communication. Leave no room for ambiguity. Ask dumb questions. Write things down clearly.
- Health is #1: Stay hydrated. Eat a balanced diet. Sleep 8 hours a night. Exercise frequently. Maintain good social and mental health. Not doing so affects your mood and long-term productivity.
- Do the right thing.

## About Composite

## Company Overview
- **One-liner**: Composite provides an AI-powered browser assistant that learns a user’s work patterns to automatically detect and complete repetitive web-based tasks across any browser.
- **Entity Type**: Private (Seed stage)
- **Headquarters**: San Francisco, California, USA
- **Founded**: 2024
- **Founders**: Yang Fan Yun (CEO) and Charlie Deane

## Core Business
- **Primary industry**: AI-powered Browser Automation / Agentic AI
- **Target customers**: B2B/Enterprise professionals in roles like recruiting, marketing, sales, and security engineering at companies such as Google, Uber, DoorDash, Tesla, Salesforce, and Reddit; also appeals to individual professionals looking to automate busywork.
- **Mission or purpose statement**: To eliminate digital grunt work by giving professionals an "autopilot" for their browser, allowing them to focus on meaningful, high-value work.

## Products & Services
- **Core Product (Composite)**: A cross-browser Chrome extension that runs AI locally on the user’s device. It monitors browser activity to learn individual workflows, then proactively suggests and automates repetitive tasks (e.g., researching candidates, moving data between tools, updating project statuses, drafting emails). It is accessed via a lightweight keyboard shortcut overlay (`Cmd+Shift+Space` or `Ctrl+Shift+Space`) that stays out of the user’s way.
- **Spotlight View**: A feature that eliminates intrusive sidebars, allowing the full screen to be maintained while Composite works in the background.
- **Personalization Engine**: A system that learns from a user’s actual work habits to detect and automate specific, repetitive tasks, rather than offering generic prompts.

## Market Standing
- **Valuation/Total Funding**: $5.6 million (Seed round as of Oct 1, 2025).
- **Key Metric**: Total seed funding of $5.6M; used by professionals at hundreds of companies including Fortune 500s.
- **Notable Investors/Partners**: Led by Nat Friedman and Daniel Gross’s firm NFDG, with participation from Menlo Ventures and Anthropic’s Anthology Fund.
- **Growth Signals**: 
    - Launched on Windows (expanding from Mac-only beta).
    - Already in use at major tech companies (Google, Uber, DoorDash, Tesla, Salesforce, Reddit).
    - Actively developing a task-scheduling feature and a better mechanism to automatically surface tasks.
    - Describes itself as an "ideal tool for professionals" with a focus on high-accuracy atomic actions.

## Competitive Advantages
- **Cross-Browser/No-Vendor Lock-in**: Unlike AI-native browsers (e.g., Perplexity’s Comet, Opera’s Neon, The Browser Company’s Dia), Composite works in any existing browser (Chrome, Edge, Firefox, Safari, Comet) without requiring users to switch, migrate data, or get IT approval.
- **Privacy-First/Local Execution**: The AI agent runs entirely on the user’s device, never sending login credentials or credit card data to remote servers. Users can define blocklists for sensitive websites and must confirm high-risk actions.
- **Low Friction/Professional Focus**: Designed for complex professional workflows (e.g., Jira management, candidate sourcing across sites) rather than simple consumer errands. Setup is a simple 0-second install via a browser extension.
- **Multi-Model AI**: Uses a combination of small, open-source models for speed and cost, plus larger vision models for complex operations, avoiding dependency on a single AI provider.

## Strategic Focus
- **Automation of Invisible Grunt Work**: The company’s core objective is to intelligently predict and automatically execute the most mind-numbing, repetitive digital tasks for professionals.
- **Proactive Task Suggestion**: Key product development is centered on a "better mechanism" to automatically surface tasks Composite can handle on the user’s behalf, moving from a reactive to a proactive assistant.
- **Recurring Task Scheduling**: Building functionality to schedule tasks for recurring usage.
- **Scaling to Windows**: A major recent milestone was the release on Windows, vastly expanding its total addressable market.
- **Enterprise Security**: Continuing to invest in safeguards like blocklists, user confirmation for sensitive actions, and opt-out data collection to appeal to enterprise customers.

## Why Work Here
- **High-Impact, Early-Stage Startup**: You would be joining a well-funded Seed-stage company backed by top-tier investors (NFDG, Menlo Ventures, Anthropic) that is already signing blue-chip enterprise customers. The team is small (1-10 people), offering significant ownership and influence.
- **Cutting-Edge AI Application**: Work at the intersection of AI agents, browser automation, and human-computer interaction. The tech stack involves running local, multi-model AI agents.
- **Traction with Top Companies**: The product is already validated by paying users at companies like Google, Uber, and Tesla, indicating strong product-market fit.
- **Founding Team**: Led by Yang Fan Yun (former PM at Uber, Stanford CS valedictorian) and Charlie Deane (former founder in server proxies).
- **Remote/Hybrid/Office**: Based in San Francisco. The specific remote/hybrid policy is not publicly detailed, but given the early stage and presence in SF, a hybrid or in-office culture is likely.
- **Culture**: Focused on eliminating the most tedious parts of work, enabling "creators of the world to put their education and skills to full use."

## Sources
1. [composite.com](https://composite.com/)
2. [techcrunch.com](https://techcrunch.com/2025/09/30/composite-gets-backing-from-nfdg-for-its-cross-browser-agent-tool/)
3. [prnewswire.com](https://www.prnewswire.com/news-releases/composite-raises-5-6m-to-liberate-professionals-from-tedious-grunt-work-302572973.html)
4. [venturebeat.com](https://venturebeat.com/technology/this-browser-based-ai-wants-to-kill-the-worst-part-of-your-job)
5. [fastaijobs.com](https://www.fastaijobs.com/companies/composite)

## Other roles at Composite

- [Founding Machine Learning Engineer](https://feeny.ai/job/founding-machine-learning-engineer-adaptive-security-new-york-hghvcyvtg4t6) — New York, NY
- [Founding Machine Learning Engineer](https://feeny.ai/job/founding-machine-learning-engineer-noia-labs-paris-883jeppt1r63) — Paris, France
- [Founding Machine Learning Engineer](https://feeny.ai/job/founding-machine-learning-engineer-wholesail-san-francisco-c8zj88rptnk5) — San Francisco, CA
- [Founding Machine Learning Engineer](https://feeny.ai/job/founding-machine-learning-engineer-hiretofu-com-toronto-s77j94q766n5) — Toronto, Canada
- [Founding Machine Learning Engineer](https://feeny.ai/job/founding-machine-learning-engineer-shepherd-san-francisco-97yaer4ye1g9) — San Francisco, CA
- [Founding Machine Learning Engineer](https://feeny.ai/job/founding-machine-learning-engineer-orbit-neuro-co-san-francisco-75vy95xjx2mw) — San Francisco, CA
- [Founding Machine Learning Engineer](https://feeny.ai/job/founding-machine-learning-engineer-trove-san-francisco-ngf2c5c44bz5) — San Francisco, CA