Forward every doctor's note, discharge letter and pill bottle to one address. A timeline forms with every medication and appointment in one place the whole family can see. Agents handle the reminders, the appointment prep and the insurance, so more time is spent on loved ones, less on paperwork.
The literature is unanimous on the causes of avoidable readmission, and they are all coordination problems that families are quietly absorbing.
The cause is almost never the medicine. It's the medication reconciliation that nobody did, the follow-up nobody booked, and the specialist letter that never reached the GP. The family caregiver, usually a daughter, usually working, is the load-bearing wall.
Dose changes after discharge get lost between the hospital, the pharmacy and the kitchen counter. Nobody is sure which version is current.
Specialist appointments arrive in letters, not calendars. Reminders go to the patient, not the caregiver who actually drives.
Specialists, GPs and family hold three different versions of the truth. Decisions get made on the worst one.
Claims stall waiting for one referral letter. Appeals take weeks. The family caregiver becomes an unpaid claims adjuster.
"The lock-in is the family, not the individual. One sibling signs Mum up. Within four weeks the brain holds more than any sibling does in their head, and the whole family is dependent." Numa thesis
Numa is a temporal knowledge graph for one patient, fed by everything the family forwards, read by agents that handle the legwork.
Discharge letters, prescription confirmations, insurance correspondence, photographed pill bottles. LLMs extract structured events: meds, doses, appointments, diagnoses, red flags.
Events become nodes in a knowledge graph that knows when each fact arrived, who said it and what it changed. Dose history, symptom trajectories, claim status, all queryable in time.
Reminders that know the dose changed last Tuesday. Night-before appointment briefs. Daily digests for distant siblings. Insurance follow-up that chases its own missing referral letter.
One chronological feed of meds taken, calls held, documents extracted. Every sibling sees the same source of truth, in the same order.
One address per patient. Every email, every photo, every PDF lands in a structured queue with extractions you can audit and accept.
With dose-change context: why Apixaban moved from 5mg to 2.5mg, and which letter said so.
Tracked end-to-end. Numa knows what's blocking and chases the referral letter so the family doesn't have to.
The night before, a brief: what's changed, three questions worth asking, what to bring.
The version James in Manchester actually reads. Editorial, calm, three minutes long.
Auto-summarised. Decisions, action items by person, red flags, transcript on demand.
Wellthy and Cariloop run human concierges at $200–$400 per family per month. They can't price below their cost structure. We do the same job with agents at a fraction.
Mid-size employer contracts (the channel Wellthy and Cariloop already opened) come in at high six figures ARR. Health plans buy on PMPM once we have readmission data; every avoided readmission saves them ~$15k. Care homes white-label per facility, weeks to close.
Medical student and researcher at UCL's Institute of Health Informatics, where he spent a year building methods to detect AI model drift and clinical data pipelines that ran against Hong Kong's full population health record.
Numa is personal: Chris cared for his late father through a complex final illness, and watched the same paperwork-vs-presence trade-off play out again with his mentors' patients' families.
Built context maps and ran agents over graph memory structures while working with the UNESCO Chair on AI. Well-known in the London university startup community.
The Numa context layer is a continuation of that work: a per-patient temporal knowledge graph designed for agents to read and reason over, not just retrieve from.
We met at a London networking event almost a year ago, found we were chasing the same idea (storing real context for agents) and won well at a 48-hour hackathon. We've shipped together since. The caregiving wedge crystallised in the last month.
Within 12 months: paying care homes, employer pilots, retention data on consumer families, and ideally one signed health-plan pilot: the category-defining signal.
Sitting down with caregivers, care home staff and clinicians. Watching them use it. Shipping the features they actually ask for.
5–10 families on the consumer plan with retention data. First care-home and employer pilots, free or near-free, to land case studies.
One signed health-plan pilot is a category-defining signal. PMPM economics kick in once we have readmission data they can underwrite.
We'd love to talk to caregivers, care-home operators, employer benefits leads and health-plan teams. Or anyone who's lived this.