Market Analysis

Existing applications tend to be polarized, focusing either exclusively on "prescribed medication" or relying heavily on "manual entry." As a result, they fail to balance ease of registration with scheduling flexibility. Furthermore, many current solutions lack robust drug interaction checks, leaving significant gaps in user safety.

Core Value

Lifestyle-Linked Scheduling

Integration with daily rhythms A scheduling architecture designed to sync medication with the user's natural daily patterns, such as mealtimes and sleep cycles.

Universal Smart Onboarding

Fast medication registration via OCR + Database A seamless onboarding experience that utilizes OCR and database integration for immediate and accurate data entry.

Safety-Guard Interaction Check

Ensuring safety through interaction alerts A core safety layer that automatically checks for potential side effects and drug interactions to ensure user well-being.

Feature 1: Lifestyle-Sync Auto-Scheduler

• Overview: Automatically generates an optimal medication schedule by integrating the user’s daily habits (mealtimes and sleep cycles) with specific medication requirements.
• Key Point: Eliminates the hassle of manual configuration by seamlessly embedding medication reminders into the user's natural daily workflow.
• Demo: [Video 1: Profile Setup] & [Video 3: Schedule & Log] Shows how routines are set and how the AI maps doses to those specific times.

Feature 2: Smart OCR Registration

• Overview: Scans prescriptions and medication packaging using the Vision API to instantly extract information and cross-reference it with the database.
• Key Point: Whether for prescribed or OTC medications, the system identifies candidates via OCR and ensures 100% accuracy through database completion and final user verification.
• **Demo: [Video 2: Add Medication] Demonstrates the OCR-to-DB flow and the two-step validation process.

Feature 3: Safety-Guard & AI Insights

• Overview: Automatically checks for potential drug interactions with existing medications during registration, providing AI-driven safety advice.
• Key Point: Ensures user safety through a robust "Registration Block" feature and personalized information powered by the Gemini API.
• **Demo: [Video 4: Notifications] & [Video 5: Medication Records] Highlights the delivery of safety alerts and the logging of intake evidence.

Data Flow Architecture

This system integrates medication information with user data to provide end-to-end management, from automated notifications to record-keeping.

• Ingestion & Normalization: Consolidates and standardizes data from Prescriptions, Packages/Labels, and the Medication DB into a unified input form.
• Attribute Integration: Syncs user profile data (age, gender, medical history) and daily habits (routines) with the core system.
• Analysis & Generation: AI generates potential warnings (interactions/side effects) and calculates a personalized "Generated Schedule" by mapping medication requirements against the user's routine.
• Notification & Execution: Dispatch notifications via server logic and manage them as calendar-based DB events.
• Logging: Saves actual medication outcomes to DB Records to maintain an immutable evidence trail.

This end-to-end design ensures data aggregation, lifestyle optimization, and reliable execution logging, strictly following the architectural diagram.

My Contribution

I designed and achieved team consensus on this end-to-end system flow. My work focused on normalizing diverse inputs and ensuring a seamless, robust transition from habit-based schedule generation to record retention.

Key Architectural Features

This system is built on a Three-Tier Architecture (Client/Server/Database) with a focus on loose coupling between functional modules.

• Service Decomposition: Segregated the server into three distinct services—Auth, Data, and Processing—to clarify component responsibilities.
• Centralized Dynamic Processing: Consolidated Google Cloud Vision API (OCR) integration, AI-driven scheduling, and notification logic within the Processing Service.
• Separation of Data Concerns: Implemented a logical separation between user-specific data (Personal DB) and reference-only medication master data (Medical/General DB).
• Server-Side Heavy Lifting: Integrated external APIs on the server side to minimize client-side processing overhead and optimize mobile performance.

My Contribution

To eliminate data inconsistency risks inherent in OCR-only registration, I designed an "API/DB-Driven Data Fetching" architecture, ensuring the system prioritizes verified medical database records over raw OCR text.

Data Model Strategy

The data architecture is structured into a four-tier system based on data roles and update frequencies, balancing high data consistency with system scalability.

• Primary DB: Manages core master data (e.g., User, Medicine) that requires the highest level of integrity and consistency.
• Secondary DB: Aggregates dynamic operational data (e.g., Schedule, Notification) that is generated or updated on a daily basis.
• Log/History DB: Segregates immutable execution records (e.g., Medication Record, Notification Log) to ensure data auditability and analytical performance.
• Auxiliary DB: Defines supplemental data (e.g., User Device, Interaction) required for specific functional logic such as push notification routing and interaction checks.

My Contribution

I spearheaded the schema design, implementing a four-tier data responsibility model to prevent resource contention between high-consistency master data and high-frequency operational updates, ensuring long-term system stability.

Logic & Flow Control

Defined the end-to-end flow control for the medication registration process, including comprehensive exception handling from input to final storage.

• Multi-Input & Fallback: Supports both camera (OCR) and manual entry, with robust error handling for OCR failures or missing Drug Identification Numbers (DIN).
• Automated Risk Detection: Implemented a safety mechanism that cross-references the Interaction DB and immediately triggers alerts if potential drug-drug interaction risks are detected.
• Data Enrichment Logic: Built an auto-fill workflow that automatically queries the Medication DB to supplement incomplete user input and populate the registration form.
• Data Integrity Assurance: Integrated a mandatory "Edit and Confirm" step for users to verify automatically extracted data, preventing incorrect records from entering the system.

My Contribution

Independently completed the entire process design and flowchart development for this feature. I spearheaded the team consensus on all logic, including complex exception handling scenarios.

Key Architectural Features

This system enables dynamic, schedule-driven notification generation and ensures reliable logging of all user medication actions.

• Scheduling Foundation: Implemented Agenda as the job scheduler. Server-side logic dynamically controls execution timing based on medication rules stored in the Schedule DB.
• Notification Status Management: Tracks delivery states within the Notification DB. Integrated with Expo Push Notifications to automate the entire lifecycle, from dispatch to logging.
• Data Lifecycle: Maintains an evidence trail by persisting notification history in the Notification Log and recording actual intake actions in the Medication Record.
• Unified Logging Design: Constructed a traceable data model that maps the entire process from notification issuance to final user response.

My Contribution

I was responsible for the overall design of the notification system and the definition of the log data models. Based on team proposals, I spearheaded the implementation of Agenda, developed the production code for Expo integration, and significantly improved system stability through extensive refactoring.

[Design Pivot] From OCR "Full Automation" to DIN-Centric "Two-Step Validation"

Fully automating data ingestion via OCR led to data inconsistencies and a poor UX due to excessive manual corrections. To prevent this, I re-engineered the architecture into a two-step structure to ensure data reliability.

• The Challenge: Data Inconsistency and Operational Burden Reliance on OCR alone resulted in unstable accuracy due to varying prescription formats. Forcing users to identify and fix OCR errors became a major barrier to retention.
• The Solution: Building Trust via DIN (Drug Identification Number)
  1. Step 1 (Automated Fetching): Limited OCR’s role to identifying the DIN, then immediately pulled verified "Master Data" from the medication database.
  2. Step 2 (Supplemental Entry): Separated variable information (e.g., dosage, instructions) into a final user-confirmation flow.
• The Outcome Balanced system-level data integrity with a significant reduction in cognitive load for the user.

My Contribution

Identified the technical limitations of OCR early in the process and proposed the "Two-Step Input Architecture" prioritizing data integrity. I spearheaded the final decision-making for the entire logic, including complex exception-handling scenarios.

• Frontend: Built with React Native (Expo) and TypeScript, ensuring robust and highly efficient cross-platform development.
• Backend: Utilized Node.js (Express) for high-speed notification processing, with Python handling advanced parsing of unstructured prescription data.
• AI & OCR: Leveraged Google Cloud Vision API for prescription analysis and the Gemini API to automate complex drug-drug interaction checks.
• Database: Combined MongoDB for its flexible data structure with Supabase (Storage) for reliable static file management.

Step 1: Enhancing & Expanding Flexible Medication Scheduling

Based on user feedback from the prototype, the goal is to improve handling of edge cases (e.g., irregular intake times) and optimize the user experience of core features.

Step 2: External Data Integration & Reliability Improvement

Transition from the internal database to official and external medical APIs. This will establish a robust information infrastructure where drug data is automatically updated to ensure maximum accuracy.

Step 3: Optimizing Drug Interaction Alert Severity

To balance safety and convenience, a feature will be introduced to customize alert severity (e.g., display only, ignore, or forced notification) based on the specific clinical risk and user context.

• Mobile Health Integration Integrates with wearable devices such as Apple Watch. Beyond immediate notifications, it leverages vital data to provide reminders at the most effective moments.
• Family Support System Centralizes medication schedules for multiple family members, including children and seniors. Features a multi-user system to monitor the adherence of family members living apart.
• AI-Powered Hyper-Personalization Delivers highly personalized guidance and customized alerts based on individual health profiles, including pregnancy, allergies, and liver or kidney function metrics.