Engineering an early-stage
care-monitoring platform
- ~7 months
- Engineering cooperation on a prototype and first commercial version.
- 2 + 1
- Two developers and one DevOps engineer from Panonit.
- Computer vision + LLM
- Activity recognition combined with LLM-based interpretation.
SUMMARY
Engineering an early-stage care-monitoring platform that combines computer vision, LLM-based interpretation and high-volume activity data processing.
CLIENT & INDUSTRY
Client & industry overview
Alvis develops an AI-powered monitoring concept intended to give families greater visibility into the daily wellbeing of older adults. The product combines camera-based visual and audio observations with analysis of relevant events, routines and changes in activity. Its intended user experience includes daily summaries and notifications that can help families understand when attention may be needed.
Potential applications in care homes and wider healthcare settings formed part of the product roadmap, with opportunities to connect monitoring insights to the wider healthcare IT ecosystem.
CHALLENGES
Business challenges & objectives
PLATFORMS & SCOPE
Products, platforms & scope of work
-
AI-POWERED CARE MONITORING APPLICATION
Frontend and backend functionality supporting observation-based monitoring and interpretation.
-
COMPUTER VISION AND AI WORKFLOW
Selective capture and analysis of relevant camera observations using OpenCV and Google Gemini.
-
ACTIVITY DATA PROCESSING
Organising large event and log volumes into patterns and time-based activity summaries.
Scope of Engineering Services & Responsibilities
- Panonit supplied additional product-engineering capacity to extend an existing codebase and support implementation across application, AI and data-processing workflows. Work was coordinated through shared tasks, implementation ownership and pull-request reviews.
- Frontend and backend application implementation and extension.
- Computer-vision-assisted selection of relevant moments for recording and analysis.
- Integration of selected observations with LLM-based interpretation.
- Data-processing logic for large observation logs and recurring activity patterns.
- Engineering support for AI API consumption estimation and processing efficiency.
- DevOps support within the three-person engagement team.
WHAT WE BUILT
New modules, functionality & core services developed
-
01
Selective observation processing
OpenCV-assisted filtering to identify relevant moments and avoid treating every frame as equally important.
-
02
LLM-based interpretation
Application functionality connecting selected monitoring observations with Google Gemini analysis.
-
03
High-volume log organisation
Algorithmic processing to organise approximately two million log entries per day.
-
04
Routine and pattern analysis
Aggregation of observations into daily, weekly and monthly activity patterns.
-
05
Reporting workflow
Product workflow designed to support daily summaries and notifications based on analysed observations.
-
06
AI consumption modelling
Estimation of request sizes and token usage against provider pricing.
SYSTEM INTEGRATIONS
System integrations & customer-specific implementations
The core integration in this engagement connected camera observations, computer-vision processing, Google Gemini interpretation and application data handling. Supabase supported the application’s data layer. These components formed a capture-to-analysis workflow within the prototype.
The product roadmap envisaged integration into typical healthcare IT ecosystems, including electronic health records (EHR/EMR), care management and care-home administration platforms, remote patient monitoring systems, nurse-call and alerting solutions, telehealth platforms, clinical communication tools, and hospital or clinic information systems.
AI-DRIVEN PRODUCT DEVELOPMENT
AI-driven product development & platform engineering
Engineering focused on building and refining an AI-enabled application: integrating computer vision with LLM interpretation, developing high-volume activity data workflows, and balancing analytical capability with processing efficiency and operating costs as product requirements evolved.
- Applied selective recording and analysis to avoid unnecessary AI requests.
- Structured large observation streams into data suitable for time-based aggregation and pattern identification.
- Supported iterative feature delivery and shared technical coordination during prototype development.
TECHNOLOGY
Technologies & engineering expertise
- FRONTEND
- TypeScript; application interface development
- BACKEND
- Python; application and processing logic
- DATABASE / PLATFORM
- Supabase
- AI / LLM
- Google Gemini; observation interpretation and API-consumption modelling
- COMPUTER VISION
- OpenCV; event selection and selective capture
- DEVELOPMENT SUPPORT
- Lovable; rapid application development support
- ENGINEERING PRACTICES
- Frontend/backend implementation, code reviews, DevOps support and iterative delivery
DOMAIN EXPERTISE
Industry & domain expertise acquired
The work combined healthcare-adjacent product design with practical AI engineering challenges in a home-monitoring context. Key domain areas included:
- Older-adult wellbeing monitoring and family-facing visibility into daily activities.
- Event detection from continuous camera observations and the distinction between raw data and actionable information.
- Longitudinal activity analysis, including recurring routines and time-based summaries.
- Trade-offs among monitoring coverage, data volume, AI request frequency and processing cost.
- Prototype validation and iterative requirements in an evolving AI product.
STANDARDS
Relevant industry standards & protocols
The project involved monitoring data and AI-driven interpretation in a care-related setting. Relevant engineering considerations include privacy-conscious handling of sensitive observations, access controls, secure application interfaces, responsible AI outputs and appropriate data retention.
THE RESULT
Results & business impact
Working AI product
Application functionality connected camera observations with selective processing and Gemini-based interpretation.
Data-processing capability
Algorithm developed to organise approximately two million daily log entries and support recurring activity patterns.
Processing efficiency
OpenCV-assisted filtering helped avoid unnecessary AI requests; API request and token costs were modelled.
Early-stage product validation
The platform was tested in home settings with six individuals during the prototype phase.
Care-monitoring potential
During testing, a monitoring alert prompted a family member to contact an individual and seek medical assistance.
ENGAGEMENT
Engagement & cooperation overview
- CLIENT
- Alvis
- INDUSTRY
- Healthcare technology / AI-powered care monitoring
- PROJECT
- Early-stage AI-powered monitoring application
- DURATION
- Approximately 7 months
- PANONIT TEAM
- 2 developers + 1 DevOps engineer
- COOPERATION MODEL
- Additional engineering capacity; shared implementation and code reviews
- CORE SCOPE
- Application engineering, computer vision, LLM integration, data processing and AI efficiency
- PROJECT STAGE
- Prototype / first commercial version
Ready to see what we can do for you?
Tell us about your platform — we’ll share how we’d modernise it without stopping your business.