Orchestration first
The system begins by orchestrating the best open models — task-aware selection, evaluation, guardrails, and governed memory. Model-agnostic by design; no single upstream dependency.
Harmoniq Labs — Founding Engineering Team
Apply your craft to the world we want to live in.
We are building HAIS — coordination intelligence that begins by orchestrating the best open models, and reaches beyond them. Democracy-aligned by design. Built for after AGI, not the race toward it.
Read the wider thesis — Full Stack Alignment, TELO–AYNI, Alive — at harmoniq.world ↗.
Mission
Harmoniq Labs builds HAIS as a decentralised coordination protocol, not a monolithic model — a complete AI system designed to align intelligence with the people it serves.
The bet is architectural, not a capability race: we begin by orchestrating the best open models, then layer in active inference, causal reasoning, and world models — coordination intelligence, built deliberately.
Phase 1 deploys on partner compute. This is a software and protocol effort: energy-aware, coordination-first, built to last a decade rather than win a quarter.
How we build
Orchestration first — and what the system grows into.
Active Inference
Planning and orchestration under uncertainty — perception–action loops, not mere prediction.
Causal AI
Reasoning about intervention and consequence, beyond correlation.
World models
Structured, durable state the system reasons over — coordination that reaches beyond language.
Open positions
Founding roles.
Mixed seniority — leads and strong individual contributors.
Orchestration
Orchestration — the core
What you'll work on
The orchestration layer at the heart of HAIS — selecting across leading open models per task, wrapping every call in evaluation, governed memory, and guardrails, so quality never depends on any single upstream provider. This is the founding seat of the engineering core.
Who you are
Strong ML engineer or researcher who has built systems that combine multiple models; you think about evaluation, routing, and reliability as first-class problems, and you are comfortable owning an architecture end-to-end.
Signals we value
Shipped multi-model or agentic systems; a feel for evaluation and quality under real workloads; model-agnostic instincts and a bias for what works.
What you'll work on
The infrastructure that runs HAIS efficiently on partner compute — inference pipelines, scheduling, throughput and cost across a heterogeneous model field. Phase 1 runs on external providers; you make that fast, cheap, and reliable.
Who you are
Strong ML-systems / infra engineer: inference pipelines, GPU orchestration, performance.
Signals we value
Real efficiency wins; comfort across heterogeneous compute and models; pragmatism over purity.
What you'll work on
The applied layer and demonstrations where HAIS acts as OS-level intelligence — task classification, orchestration, and scheduling made tangible and usable.
Who you are
Full-stack or applied-ML engineer who can turn capability into a working, legible product surface.
Signals we value
A portfolio of things you've built and shipped; taste in interface and demonstration; speed without sloppiness.
What you'll work on
The protocol layer that makes HAIS a decentralised coordination system rather than a single monolith — interfaces, incentives, and the coordination model across independent compute providers.
Who you are
Distributed-systems depth; ideally exposure to decentralised protocols or coordination mechanisms. You care about incentive design as much as code.
Signals we value
Shipped distributed systems at scale; an instinct for protocol design; comfort at the boundary of systems and mechanism.
Coordination intelligence
Coordination intelligence
What you'll work on
Bringing active-inference principles into how the system plans and acts under uncertainty — a coordination mechanism that reasons about its own uncertainty, not just predicts.
Who you are
Strong grounding in reinforcement learning, planning, world models, or active inference; able to move between theory and a running system.
Signals we value
Practical experience turning principled agent frameworks into deployed behaviour. Familiarity with the active-inference literature is a plus, not a requirement.
What you'll work on
Causal representation learning and inference — giving the system the ability to reason about interventions, counterfactuals, and consequences rather than surface correlation.
Who you are
Research or applied background in causal inference / causal representation learning; rigorous, and comfortable with messy real-world data.
Signals we value
Work that bridges causal theory and applied systems.
What you'll work on
The structured world-model state and durable, retrievable memory the system reasons over — coordination that reaches beyond language, and memory that persists and learns without forgetting.
Who you are
Experience with world models, memory-augmented systems, retrieval, or structured state; comfortable making memory and state first-class components.
Signals we value
A track record of making state and memory load-bearing rather than a bolt-on.
Alignment
Alignment — the binding discipline
What you'll work on
The alignment and evaluation practice for HAIS — turning “democracy-aligned” from principle into measurement: evaluation suites, interpretability, guardrails, and the standards for what we will and will not build.
Who you are
Alignment, safety, evaluation, or interpretability background; rigorous, and able to hold a values position without hand-waving.
Signals we value
Concrete evaluation or interpretability work; an ability to operationalise values into tests; seriousness about the societal stakes without cynicism.
Who we're looking for
For researchers and engineers who no longer wish to pour world-class skill into extraction — but into coordination, alignment, and a civilisation worth living in.
We hire for depth and judgement, not for pedigree signalling. A small founding team, high autonomy, real ownership — the opposite of a layer in a thousand-person org.
Democracy-aligned is a working discipline here, not a tagline: it shows up in what we evaluate and what we refuse to build. We value people who can hold a hard problem for a decade, not a sprint.
How we hire
Founding-team mode, described plainly.
Equity-first compensation, with cash calibrated to stage. Honest about the trade: meaningful ownership, early risk.
Remote-friendly, hubbed in Zürich. Relocation and visa support available where useful.
We read every application ourselves.
- 01
Intro conversation
A short call to understand what you want to build, and what we're building.
- 02
Technical & architecture discussion
A deep conversation grounded in real problems — no adversarial puzzle theatre.
- 03
Meet the team
Working sessions with the people you'd build alongside.
- 04
Offer
Honest terms, quickly. We read every application ourselves.
Apply
Write to us.
A short, considered note beats a polished CV. Tell us where your craft belongs.