CareForge — clinical safety at the point of care
Putting swallowing-safety triage in the hands of the ~640,000 frontline carers who actually decide whether a frail elder's meal is safe — in their own language, on the device they already carry, with no clinician needed in the room.
Aspiration pneumonia is a leading preventable cause of death and hospitalisation in frail elderly. The people who actually decide meal safety — foreign domestic helpers, family caregivers, frontline care-home staff — have no screening tool, no training in their first language, and no clinician on hand at meal time. Hong Kong's care-food standard exists. Detection at the point of care does not. This is a structural reach gap of mandate, language and channel — not a capability criticism of any operator. This is complementary infrastructure to HKCSS's eight years of 照護食 sector work — extending point-of-care reach where the policy and standard already exist.
Preventable
Aspiration pneumonia — a leading preventable cause of hospitalisation in frail elderly. Tens of thousands of Hong Kong's frail elderly residents face meal-time aspiration risk every day.
~640,000
Frontline carers in Hong Kong deciding meal safety today — mostly with no screening tool in any language they read.
Structural
Reach gap of mandate, language and channel — the standard exists; the point-of-care tool does not.
Three operational layers built around a public clinical resource. Each layer addresses a specific human moment at the point of care — the meal being plated, the cough after a sip, the carer who needs an answer in the next thirty seconds.
A frontline carer photographs the prepared meal. On-device computer vision checks the texture against the IDDSI framework (International Dysphagia Diet Standardisation Initiative — the global care-food safety standard) and tells the carer, in their own language, whether the dish is safe for that resident. Live and deployed today — not a promise.
Non-invasive acoustic analysis of swallow and cough sounds, surfacing a conservative aspiration-risk flag. Edge-first, privacy-preserving, no clinician required to capture the signal. Research-grade and staged — see Clinical safety below for thresholds and escalation rules.
Everything is delivered in the languages real frontline carers actually speak: Cantonese, English, Tagalog and Bahasa Indonesia. This is the channel through which clinical safety reaches the bedside — explicitly the delivery layer, not the product itself.
Co-governed public clinical resource
Deployment generates a longitudinal, multilingual, Asian-population swallowing-safety dataset. It is held as a co-governed public clinical resource — stewarded jointly by the HKU Swallowing Research Laboratory and an independent registered s.88 charity — not as anyone's proprietary asset. The legacy is a Hong Kong public good, not a private moat.
CareForge is not a one-person project. The clinical-safety spine is anchored at HKU. The data steward is an independent registered charity. The delivery team is funded by the impact grant we are seeking — recruited on award, not pre-existing free labour.
Raymond Chau
Founder & Principal Investigator, Carewells Limited
Carries the operational responsibility for CareForge through Carewells Limited — a Hong Kong social enterprise and recipient of the HKSAR Government TechUp grant (HKD 600,000, 2025) for AI-assisted swallowing-assessment technology. Carewells is the operating vehicle for the service; clinical authority sits with the HKU Swallowing Research Laboratory, not with the founder.
Karen Chan
Clinical Advisor & Clinical-Governance Anchor — HKU Swallowing Research Laboratory
Owns the clinical-safety protocol, validation methodology and safety sign-off for CareForge. Advisor since founding. Her involvement is what converts "carer-run swallowing screening is hard" from a risk into a credibility strength.
Clinical-AI / machine-learning lead · acoustic voice-biomarker researcher · multilingual content & field officers · care-home deployment coordinators · programme manager. The grant funds the team. The team is not assumed.
The screening dataset is co-stewarded by the HKU Swallowing Research Laboratory and Full Linkage Community Care Association — a separate legal entity from Carewells, never described as linked to Carewells. Full Linkage Community Care Association: Registered s.88 charity dedicated to community care in Hong Kong.
Operating entity Carewells Limited is a Hong Kong social enterprise — not a s.88 charity. The two roles are deliberately separated so the public asset cannot be privatised.
Public impact money should not build a private moat. CareForge is structured so the public asset stays public — by design and in writing, not as a marketing claim.
We do NOT take ownership of the screening dataset. It is co-governed by the HKU Swallowing Research Laboratory and an independent registered s.88 charity, under a published data-ethics and consent framework.
We do NOT keep grant-funded models as proprietary IP. Models trained on the consented public dataset are released under the same co-governed framework, with open methodology.
We do NOT route public impact funding into a private moat. Where commercial activity exists, it sits in a separate, named layer — enterprise tooling, certification and B2B deployment — beside the public asset, never on top of it.
We do NOT replace clinical judgement. Every CareForge signal is triage to a clinician — never diagnosis, never autonomous medical advice.
We do NOT operationalise without HKU clinical governance. Validation, conservative thresholds and safety sign-off live with the HKU Swallowing Research Laboratory, not with us.
Carewells' commercial work is built on the layer beside the public asset — enterprise screening deployment, care-food certification and compliance tooling, B2B institutional contracts and downstream services. It is never built on owning the grant-funded data. The separation is stated plainly so judges, operators and families can verify it.
Carer-run swallowing screening is a real intervention with a real failure mode. We have addressed it in the design of the product, not in the marketing.
CareForge surfaces a risk flag and a recommended next step. It does not diagnose, does not prescribe, does not replace a speech therapist or a clinician. The output is always either "this needs professional assessment" or "this is within typical bounds — continue, observe, and re-check at this interval."
Uncertain results always escalate to "seek professional assessment". Thresholds are deliberately tuned so the failure mode is over-referral, never a missed risk.
CareForge is positioned plainly in the wellness and triage tier, not as a regulated diagnostic device. The boundary is defined in the product, in the consent flow, and on this page — not buried in a footnote.
Clinical validation, threshold-setting and safety sign-off live with Karen Chan and the HKU Swallowing Research Laboratory. The protocol is theirs, not ours.
90-day RCHE pilot — one-pager
A 90-day pilot designed to be safe to say yes to. Short, clinically supervised, no procurement lock-in, and the operating data stays with the home.
Duration
90 days, with a clinical mid-point review at day 45
Cohort
A single care-home site; up to ~30 residents flagged by the home as appropriate
Cost to the home
Nil. Tooling, set-up, training and clinical supervision are at our cost during the pilot.
Three plain measures, agreed up front. (1) Did frontline staff use it during at least 60% of high-risk meals across the 90 days — not just in training? (2) Did the home see a measurable reduction in near-miss meal-safety incidents at the mid-point review compared to the prior 90 days? (3) Did the home commit, in writing, to keeping the tool live for at least six months post-pilot? No vanity metrics.
If you run a care home, advise on swallowing or care-food safety, fund preventive eldercare, or want to help steward this work as a public good — we want to hear from you. Replies typically within two working days.
Prefer email? Write to info@seniordeli.com