AI Engineer

Design and build the intelligence at the core of Haven — local inference, retrieval, memory, planning, reasoning, and evaluation that make a personal AI useful on the Mac someone already owns.

The role

Haven is a local personal intelligence system built around AIOS, our neurosymbolic architecture. This role works across the learned and deterministic parts of that system — improving the paths that understand a request, retrieve knowledge, plan, reason, generate, and verify.

Responsibilities

• Design and implement core AI paths across conversation, retrieval, personal knowledge, planning, reasoning, and verification.
• Build reproducible evaluations for model and system behavior, then use failure data to decide what should change next.
• Improve local inference on Apple silicon, balancing model quality, latency, memory use, context length, and the limits of personal hardware.
• Work across learned and deterministic components of AIOS, using language models where they help and structured systems where they are more reliable.
• Investigate failures in real turns — wrong answers, missed context, poor retrieval, brittle planning, or bad tool use — and ship concrete fixes.
• Partner with product and platform engineers so improvements in intelligence become dependable product behavior, not benchmark wins or demos.
• Write clearly about what you built and why so the team can examine, challenge, and improve it.

Qualifications

• Strong Python and hands-on experience building production or near-production AI systems, not only training runs or notebooks.
• Experience with retrieval, evaluation, memory, reasoning, planning, or tool-using language-model systems.
• Comfort with local or on-device inference and the practical constraints of model size, memory, latency, and context.
• Ability to debug the whole intelligence path: classification, retrieval, context, model output, tools, and the software around them.
• Interest in systems that combine statistical models with deterministic software rather than treating the language model as the entire architecture.
• Clear written communication and a habit of making technical work inspectable by other engineers.
• Strong ownership and comfort with ambiguous problems on a small research and product team.

Also hiring

Build and operate the end-to-end personal intelligence system, from model packaging and deployment to observability, reliability, and continuous improvement in production.