Heidi Cut 108 as Its CTO Chose Robotics. Pitch the FHIR Engineers First

If you hire clinical AI engineers in Australia, one of the country's deepest pools of that talent has just come onto the market, and you won't be the only one chasing it. Heidi Health, the Melbourne AI scribe unicorn, is cutting 108 people, or 18% of staff. The Melbourne R&D team is expected to take the heaviest hit, and the cut landed in the same week CTO Yu Liu left to pursue robotics. You're now competing with a former boss who knows exactly who his best people were. He won't want all of them, though, and that tells you where to start.

What happened at Heidi Health, and why the timing matters

Heidi is laying off 108 people, about 18% of a workforce of more than 600, and two co-founders are leaving at the same time. The cut reverses a growth plan announced only weeks earlier.

The facts that matter for sourcing:

• Who is leaving: CTO Yu Liu and CFO Waleed Mussa. CEO Dr Thomas Kelly told staff about the cuts.
• Where it hits: The AFR reported the cuts would fall heavily on Melbourne R&D, plus the large support and sales teams. Some roles are moving offshore.
• The money: In September Heidi raised US$340 million ($475 million) in a round led by Blackbird at a $1.26 billion valuation. That total included a US$240 million facility from General Catalyst, which is repaid from the revenue it generates. Debt repaid from revenue explains a lot of the cost pressure.
• The reversal: About six weeks before the cut, Kelly told Bloomberg that Heidi planned to add as many as 150 staff over the next year.

Kelly was blunt about the leadership change: "The shape of a technical leader we need now is not what we needed at the beginning." By his account (reported through the AFR, so not Liu's own words), Liu wants to build something from the ground up in robotics.

When a CTO walks out in the same week as a layoff, he becomes the best-informed recruiter in the market. He knows who shipped, who was underused and who would follow him. Outside hirers have one structural advantage: he is building from the ground up, so he has no team for them to join yet.

Will Yu Liu recruit his old team? Only part of it

Liu's pull will mostly reach Heidi's model, speech and hardware engineers. The EHR and FHIR integration engineers have no obvious place at a robotics startup, which makes them the pool outside healthtech hirers can win most easily.

Here's why. Liu is moving into an advisory role at Heidi focused on "models and hardware" through the end of March 2027. That brief overlaps closely with what a robotics company needs:

• In-house models. Heidi built its own models instead of renting frontier ones. It says transcription and note generation, which use most of its compute, now run almost entirely on in-house technology.
• On-device processing. Heidi launched Heidi Remote, a 21-gram clinical wearable, as part of a push to run processing on the device.

A robotics founder needs ML, embedded and on-device people, and Heidi has both kinds. He has much less use for the engineer whose job was making Heidi's notes land cleanly in a GP's practice management system.

Any non-solicit terms between Liu and Heidi aren't public, so don't build your plan on guesses about them. Plan as if he can hire whenever he wants. Either way the takeaway holds: expect to compete with him for ASR and model engineers, and expect to compete mostly with other healthtech companies for integration engineers.

Why searching "Heidi Health" finds almost nobody

A search on the company name reaches only about 3% of Heidi's staff. In Refolk's index of Australian professional profiles, 55 headlines mention "Heidi Health", and only 20 of those people list Heidi as their current employer. Heidi has more than 600 staff.

~3%

The reason is simple: startup engineers write their headline around their role, not their employer. "Staff Software Engineer" tells another engineer more than a company name does.

The 55 profiles that do mention Heidi split like this:

• Melbourne: 12 (8 Melbourne, 4 Greater Melbourne)
• Sydney: 10
• Titles in the sample: CTO & Co-Founder, Engineering Lead, Staff Software Engineer, Tech Lead for Mobile Foundations

Sydney nearly matches Melbourne, even though the press framed this as a Melbourne R&D story. Don't filter by city too early.

A Boolean search on employer name will miss most of the people you want. Refolk handles this differently: you describe the work ("built FHIR integrations for an Australian AI scribe") and get back people whose profiles, repos and posts show that work, whatever they put in their headline.

The three skill pools to search instead

Search on what Heidi engineers built, not where they worked. That work falls into three pools: EHR integration, speech and clinical NLP, and on-device or hardware engineering.

Two terms first. FHIR is the interoperability standard integration engineers use to move health records between systems. Clinical NLP is language processing applied to medical text, such as turning a consult transcript into a structured clinical note.

Here's how deep each pool runs in Refolk's index, pulled in November 2026:

Pool 1: EHR and practice-software integration

These are the engineers to approach first, because Liu is unlikely to want them. Search for:

• FHIR, HL7
• Best Practice, MedicalDirector, Cliniko, Halaxy, MediRecords
• Practice management integration

Integration work at Heidi got harder just before the cut. Less than a week earlier, Best Practice, the most-used GP software in Australia, said running Heidi's newest product breaches its licence. An engineer who has built against Bp under those conditions has something few FHIR resumes show: experience shipping an integration while the vendor is pushing back.

Pool 2: Speech, ASR and clinical NLP

This is the contested pool. Search for:

• ASR, speech recognition
• Clinical summarisation, medical note generation
• In-house model training

Pool 3: On-device and hardware

These engineers are the most likely to follow Liu. Search for:

• On-device inference
• Embedded, wearable
• Anyone connected to the Heidi Remote launch

Also add "Oscer", Heidi's former name. Early employees often list the old name and never update it.

Clinical NLP is the scarcest pool in the country

Clinical NLP is the hardest of the three skills to hire in Australia. Refolk's index finds only 16 profiles in Australian healthcare that list NLP, against 50 for FHIR. That's roughly 3.1 times as many integration engineers as clinical NLP engineers (the FHIR count covers all industries, so the real gap within healthcare is probably wider).

16

So Heidi's transcription and note-generation team was probably one of the largest groups with this skill in Australia. When even a handful of these people enter the market in the same week, that's a meaningful share of the national supply. Hospital networks, Harrison.ai and every AI scribe competitor will want them.

If you're trying to hire AI medical scribe engineers with real clinical NLP depth, you're in a short, crowded race for a very small pool. In that race, sending cold messages one at a time to a list that's 97% invisible to keyword search uses up weeks you don't have.

The FHIR pool is less scarce but more spread out. Telstra Health is the top employer in the index, with only 2 profiles. The rest sit at eHealth NSW, CSIRO, Deloitte and a long tail of small teams, including one titled "Team Leader Health Data and FHIR". Melbourne and the Sydney area account for 6 each. Heidi's integration engineers will be among the few in that pool who have also shipped an AI product at startup speed.

How long is the window? Shorter than Liu's launch date

The real hiring window is much shorter than the roughly five months until Liu's advisory role ends in March 2027. Several forces are competing for the same people at once:

1. Heidi's own placement offer. The statement read: "The people affected have helped build Heidi, and we'll use our network to help them find their next role, including within our portfolio." The source doesn't make clear who is speaking, and it may be Blackbird. Either way, the investor's portfolio companies get the first introductions.
2. Canva's restructure. Canva announced a cut affecting 22% of its workforce in the same week. That adds competing supply and pulls recruiter attention toward a much bigger name.
3. Direct competitors. Lyrebird Health, a Melbourne AI scribe focused on GPs and partnered with GPRA, has been advertising engineering roles such as Engineering Manager, ML Evaluation Platform.
4. Liu himself. He is building from the ground up, so he has no company to hire into yet, but he already knows who he wants.

The order matters. Portfolio introductions and direct competitors can move fast. A startup built from scratch takes longer. Outside hirers who wait for "the dust to settle" will reach engineers who already have offers.

How to write the first message

Lead with the specific work, say outright that the cut was about cost and location, and name the problem you'd hire them to solve. Generic "saw you were affected" messages will be lost among dozens of others.

Some roles are moving offshore, so part of this cut was about cost and geography, not individual performance. Engineers notice whether your outreach treats them like a discount. A strong first message does four things:

• Names the work, not the company. "You built the Best Practice integration path" beats "I saw you were at Heidi."
• Says the cut wasn't about them. One plain sentence about the offshore move is enough.
• Matches the right pool. Offer integration engineers a hard interoperability problem. Offer speech and model engineers ownership of a model, which is what Liu will offer too.
• Gives a reason to move now. A start date, a named hiring manager, a short process.

Don't criticise Heidi or speculate about the founders. These engineers just watched the company they built go through a hard week, and many still respect it. A recruiter who sounds like a competitor gloating loses the reply.

FAQ

How many people did Heidi Health lay off?

Heidi Health is cutting 108 roles, about 18% of a workforce of more than 600. The cuts were expected to fall hardest on the Melbourne R&D team and on the support and sales teams, with some roles moving offshore. Nobody has confirmed how many of the 108 roles are in Australia.

Is Yu Liu hiring for a robotics startup?

Kelly said Liu wants to build something from the ground up in robotics, which means there is no team to join yet. Liu stays on as an adviser to Heidi on models and hardware until the end of March 2027. Assume he'll go after speech, model and on-device engineers rather than EHR integration specialists.

What should I search for besides "Heidi Health"?

Search for skills and systems: FHIR, HL7, Best Practice, MedicalDirector, Cliniko, Halaxy, MediRecords, ASR, clinical summarisation, on-device inference and the company's former name, Oscer. In Refolk's index, only about 3% of Heidi's reported headcount names the company in their headline, so employer-name searches miss nearly everyone.

Why is clinical NLP engineer sourcing so hard in Australia?

The pool is tiny. Refolk's index finds only 16 Australian healthcare profiles that list NLP, compared with 50 for FHIR across all industries. Heidi's in-house transcription and note-generation work made it one of the biggest employers of this skill, which is why its departing engineers will get several offers quickly.