Careers
Work on the bottleneck of the decade.
Advanced packaging decides how much AI compute exists. Almost nobody is building software for it. If that combination interests you more than it frightens you, we should talk.
The work
Physics, not dashboards
This is not a company where the hard problem is the front end. The hard problems here are volumetric defect detection at line rate, closed-loop control at sub-micron alignment, simulation fast enough to gate a write-back, and a governance model that a semiconductor quality organisation will actually sign.
It is also unusually concrete work. A model improvement is not an engagement metric; it is a package that ships instead of being scrapped, on a line where the material is worth more than the tool that damaged it.
The constraint that shapes everything is trust. We are asking manufacturers to let software write to equipment that costs millions and produces material worth more. Every engineering decision — grounding, envelopes, twin validation, immutable audit, revocable autonomy — exists to make that request reasonable.
- Perception on real volumetric and surface inspection data
- Control and optimisation against physical processes
- Fab-edge systems where latency and failure semantics are load-bearing
- Governance treated as a first-class engineering surface
How we work
Six things that are true here
Closest to the line wins
Arguments are settled by data from a real line, not by seniority. The engineer who has spoken to the process owner most recently usually has the strongest claim.
Small team, large surface
Early-stage means owning more than your title suggests and shipping without a committee. It also means genuine ambiguity, which suits some people and not others.
Honesty as a discipline
We publish our limitations, mark aspirational figures, and document pilots that miss their metric. Internally, the standard is higher, not lower.
Domain respect
Process engineers are the experts. Our job is to encode their judgement faithfully, not to assume software people know better because the tooling is newer.
Boring where it matters
Idempotency, versioning, rollback and audit are not unglamorous chores here. They are the reason anyone lets us near a bonder.
Written thinking
Decisions of consequence get written down with their reasoning and their risks, so that the next person can disagree with the argument rather than the conclusion.
Open roles
Where we need people
-
01
Computer vision engineer — volumetric inspection
X-ray, CT and SAM perception at line rate. Segmentation and classification on 3D volumes, fine-tuning on sparse labelled defect data, and latency budgets that are not negotiable.
-
02
Control and optimisation engineer
Closed-loop control of bonding force, temperature and time profiles; stack sequencing and TSV planning as constrained optimisation; safety and envelope semantics.
-
03
Process engineer in residence — advanced packaging
Bonding, stacking, molding, underfill and warpage expertise from an OSAT, IDM or equipment vendor. You are the specification, and you will be treated that way.
-
04
Fab-edge platform engineer
Rust or C++ runtime on fab-edge hardware, Kubernetes at the edge, SECS-GEM and OPC-UA connectors, and failure semantics that hold on a production line.
-
05
Applied AI engineer — agents and evaluation
Grounded agent orchestration, retrieval with enforced citations, evaluation harnesses and regression gates that treat model drift as a build failure.
-
06
Founding go-to-market — semiconductor
Design-partner recruitment inside OSATs, IDMs and foundries. Long, technical, multi-stakeholder cycles where credibility is the entire product.
What we offer
Plainly stated
- Ownership
- Meaningful equity, and a scope of responsibility that would take years to reach elsewhere.
- Proximity to the problem
- Time on real lines with real process engineers. This is not a domain that can be learned from a dataset.
- Remote-friendly, fab-connected
- Distributed work with deliberate time in fabs and with design partners, because the line is where the truth is.
- Learning budget
- Conferences, standards bodies and courses in packaging, reliability and applied AI.
- What we cannot offer
- Large-company stability, a mature process, or certainty. If those matter to you right now, this is genuinely not the right time to join.
Hiring
How the process runs
-
Conversation
Forty-five minutes on the problem, not on your CV. You should leave understanding the domain better regardless of outcome.
-
Technical depth
A real problem from our actual backlog, discussed the way we would discuss it internally. No puzzles, no whiteboard trivia.
-
Paid work sample
A scoped, paid piece of work over a few days for most roles. We pay for it because your time has value and because it is the most predictive signal either side gets.
-
Team and references
Meet the people you would work with, ask them what is hard here, and speak to whoever you like about us.
-
Decision
Within a week of the final conversation, with reasons. Including a no.
Yield engineer, twenty-year careerAdvanced packagingI spent nine years learning to read an X-ray of a bump array. Nobody has ever asked me to write it down.
Composite drawn from design-partner and industry conversations. Illustrative, not a customer endorsement.
Why now
The case for joining early
- $30B Market being reshaped by the transition to advanced packaging.
- $7.5B Segment addressable by an autonomy layer, with no established owner.
- 0 Vendor-neutral autonomy platforms operating across the whole packaging line today.
- Pre-launch The stage at which the architecture, the culture and the equity are all still being set.
Candidate questions
What people ask us
-
For engineering roles, yes — depth in vision, control, systems or applied AI transfers, and we pair new engineers with process expertise deliberately. For the process engineer and go-to-market roles, domain experience is the requirement.
-
Funding status is discussed directly during the process rather than published. You will get a straight answer before you would need to make a decision.
-
For engineers, periodic — enough to be in fabs and with design partners rather than to live in them. For go-to-market and process roles, substantially more.
-
You will have spent a year or two doing genuinely difficult work in a domain almost nobody understands, which is a rare and durable position to be in. We would rather say that than pretend the risk is not real.
Get in touch
Tell us what you would build first.
No cover letter. Send us the problem on this site you find most interesting and what you would do about it in your first month.