VP of Research, Machine Learning

About the Company

There are over 5 billion users using basic applications today such as email, notes, and tasks that are not AI-native. Our mission is to build a proactive smart assistant for everyday users to bring intelligence to conversations, errands, organizing, and workflows, with minimal prompting. Our product focuses on achieving high reliability for long-running workflows, persistent context, and real-world task completion. The system must handle multi-step reasoning, interact with external tools, and remain reliable despite non-deterministic model behavior. Our objective is to help users complete tasks daily in an enjoyable way with over ~90% reduced time.

Responsibilities

• Set and evolve the research direction for A1’s core intelligence, including context representation, memory, reasoning, planning, and orchestration.
• Decide when to design new model architectures versus adapting or leveraging frontier open-source or commercial models.
• Define evaluation frameworks that measure real-world usefulness, robustness, safety, and long-term behavior – not benchmark vanity.
• Own alignment, safety, and guardrail strategy as first-class product concerns.
• Guide exploration of frontier techniques such as:
  • Retrieval-augmented training
  • Mixture-of-experts
  • Distillation
  • Multi-agent orchestration
  • Multimodal systems
• Shape early product intelligence direction in close partnership with product and application engineering.
• Set the technical bar for research rigor, judgment, and taste across the organization.

Requirements

• Deep experience building or evolving real machine learning systems used in production.
• Strong technical judgment around model behavior, failure modes, and long-horizon trade-offs.
• A builder’s mindset: you care about systems that work in the real world, not just ideas.
• Comfortable making irreversible or high-impact decisions with incomplete information.
• Obsession with evaluation, correctness, and how systems behave over time.
• High ownership mentality — you operate as a founder, not a manager.
• If you are looking to focus primarily on publishing, incremental benchmarks, or managing a large research organization, this role will not be a fit.

Tech Stack

• Python
• PyTorch / JAX
• GPU-based training and inference system

How We Work

The best products today in the world were built by small, world-class teams. We are a high talent density and hands-on team. We make decisions collectively, move at rapid speed, striking a balance between shipping high-quality work and learning. Joining our team requires the ability to bring structure, exercise judgment, and execute independently.