Medication non-adherence is one of the most persistent and costly challenges facing modern healthcare systems. In the UK, it costs the NHS an estimated £500 million every year and contributes to between five and ten percent of avoidable hospital admissions. With over one billion prescriptions dispensed annually to 67 million patients, the problem is systemic — and the technology deployed to address it has, until recently, been designed around an assumption that real-world care environments routinely violate.
Varun Mishra, CTO and Co-Founder of UK HealthTech venture Adhicine Solutions Ltd, has spent the past two years building an AI-powered medication adherence platform specifically engineered for the conditions patients actually live in, rather than the conditions engineers design for. Drawing on 20 years of large-scale infrastructure experience at organisations including ByteDance, LinkedIn, Lazada (Alibaba), and SunGard Global Trading, Mishra identified connectivity as the fundamental design flaw undermining every connected health solution that came before.
“The patients who need continuous AI-powered monitoring most are, as a demographic, the patients least likely to have stable internet connectivity. Every connected health solution before Adhicine was architected for the exception, not the reality.”
The core innovation is UK Patent GB2520176.5, filed with the UK Intellectual Property Office for a dual-acknowledgement offline-tolerant synchronisation protocol for IoMT devices. Under this protocol, every medication adherence event is written to non-volatile memory on the device at the moment of detection — before any network communication takes place. Each event is assigned a unique identifier, a monotonic sequence number, and a checksum. When connectivity is available, events are transmitted to the cloud backend in sequence order. The server acknowledges each event individually; the device dequeues only on receipt of that specific acknowledgement. A server-side deduplication index prevents duplicate records regardless of retry behaviour — producing a clinically reliable, ordered adherence record that survives the connectivity profile of the real home environment.
“The adherence record is produced when the patient acts — not when the cloud receives it. That distinction is the difference between a system that works in a care home and one that silently fails.”
Built on top of the synchronisation protocol are three machine learning models that run entirely on the Adhicine device, without requiring an internet connection at inference time. The first, a Long Short-Term Memory (LSTM) model, learns each individual patient’s real adherence behaviour and dynamically adapts reminder timing to match their patterns, replacing static schedules with individually calibrated prompts. Adhicine estimates a 35 percent improvement in adherence-intervention accuracy compared to fixed-schedule baselines.
The second model, a Random Forest classifier, ingests real-time IoMT telemetry and generates a predictive risk signal up to four hours before a likely missed dose occurs. Caregivers and clinical pharmacists receive a pre-emptive alert before non-adherence happens — shifting clinical care from reactive to predictive. A third Autoencoder anomaly detection model distinguishes genuine missed-dose events from sensor noise and device errors, reducing false-positive clinical alerts and improving data quality for healthcare professionals.
“We wanted clinical intelligence that worked in the real conditions of people’s homes. That meant running all three models on the device itself, so the AI remains fully functional regardless of whether the Wi-Fi is up.”
Adhicine has moved beyond proof of concept. First commercial revenue of approximately USD 24,000 has been generated through 400 units shipped via distribution partner OneRetailWorld, validating product-market fit within 12 months of incorporation. An enterprise AI and ML contract valued at £50,000 has been secured with AINAP Ltd UK. The platform holds UK Trade Mark UK00004269697 for ADHICINE SOLUTIONS (Classes 9 and 10: Software as a Medical Device, health monitoring software, medical dosage dispensers) and was independently endorsed under the UK Innovator Founder Visa scheme — assessed by a Home Office-approved endorsing body as innovative, scalable, and viable.
The platform is architecturally aligned to NHS FHIR integration standards, enabling adherence data to flow into GP and pharmacy management systems, and is on a regulatory roadmap toward MHRA and UKCA Software as a Medical Device certification. In February 2026, Mishra published three HealthTech white papers for NHS commissioners, care home operators, and HealthTech investors — contributing to the evidence base for AI-augmented adherence monitoring across NHS settings.
Mishra’s approach to connected health is informed by two decades of building reliable systems under real operating conditions. At ByteDance’s EMEA edge estate, he designed a first-of-its-kind intelligent global power matrix delivering £700,000 in contracted power savings, and a Zero Touch automation framework that reduced site delivery timeframes by 30 percent and eliminated 40 percent of manual configuration effort. A bespoke optics stress-testing tool he built independently delivered £300,000 in avoided hardware replacement costs. His total measurable impact at ByteDance exceeds £1 million. At LinkedIn, a self-initiated smart power programme delivered a 20 percent reduction in over-committed power costs and was recognised with the Certificate of Excellence, Best Employee, and STAR Performance Award.
It is this background — building infrastructure that works under real operating conditions at global scale — that shapes the engineering principles embedded in Adhicine. The same design discipline that makes infrastructure resilient at the scale of a billion-user platform translates directly to making clinical AI resilient in a care home bedroom with unreliable broadband.
“The future of healthcare AI is not in larger models or faster cloud inference. It is in clinical intelligence that works where patients actually live. The engineering challenge is not the algorithm. It is the infrastructure around it.”
As the NHS continues to seek scalable digital solutions for the chronic problem of medication non-adherence, and as connected health technology moves from controlled clinical environments into real homes, the design principles Varun Mishra has applied at Adhicine — offline-first architecture, on-device edge inference, and reliable IoMT synchronisation — represent a meaningful advance in what AI-powered clinical monitoring can reliably deliver.