ANDROID, AI, HEALTHCARE, RESEARCH

DrugQ

Research Period: 2025.07 - Present

Current Status

100%
  1. Concept
  2. Prototype
  3. Core Build
  4. Validation
  5. Output
Platform AndroidMobile
Research mHealthHealthcareHCI
Technical Multimodal AIVoice InputImage InputPrompt Design
Output Research PrototypeAMIA 2026AcceptedManuscript in Progress

Project-linked research

Publications & Research Outputs

3 outputs

Understanding Over-the-Counter Medication Decision-Making Among Older Adults and Caregivers

Li, M.; Kommaraju, H.; Lee, J. G. W.; Bae, S.; Ao, S. A.; Lee, B.; & Choe, E. K.

Accepted AMIA 2026 Symposium

Empowering Older Adults and Caregivers to Make Safe Nonprescription Medication Decisions

Li, M.; Kommaraju, H.; Bae, S.; Lee, J.; Ao, S. A.; Cox, J.; Lee, B.; & Choe, E. K.

Presentation a2 National Symposium

DrugQ: Multimodal OTC Drug Safety Assistant

Prototype-focused paper planned around multimodal input, safety verdict design, and prompt-based medication guidance.

Manuscript in progress Target venue TBD

Overview

DrugQ is an Android research prototype for a multimodal OTC drug safety assistant. It supports voice and image inputs to help users make safer medication decisions, then returns an easy-to-understand safety verdict with an explanation.

Scope

  • Implemented as a working Android prototype (not only a concept).
  • Initial medication set: Tylenol, Aspirin, Benadryl, and Rolaids.
  • Primary symptom contexts: common everyday cases such as cold/flu-like symptoms and allergies.

Problem

  • OTC medication guidance is often unclear for non-experts.
  • Users may not know how to interpret warnings, contraindications, or drug interactions.
  • When users feel unwell, they need a fast, low-effort way to get a safety-oriented answer.

System Design

Inputs

  • Voice: users speak the drug name and describe symptoms. Voice is transcribed via OpenAI Whisper.
  • Image: users capture photos of the medication (package/label or pill). The image is processed using Gemini + prompt-based interpretation.

Reasoning and Decision

  • The system uses an LLM-only approach (no separate rule engine) with prompt design to produce a final safety verdict.

Outputs

  • The assistant returns one of three outcomes:
    • Do not use
    • Consult a doctor
    • Safe for use
  • Each outcome includes a short, user-facing explanation based on the study’s questionnaire-driven criteria.

What I Built / Contributed

  • Built the full Android app end-to-end using Kotlin + Jetpack Compose.
  • Implemented the multimodal pipeline (voice + image) and integrated the model calls.
  • Designed and iterated prompts to make outputs consistent with the three-category verdict system.
  • Conducted AI integration testing and refined the interaction flow based on lab feedback.

Tech Stack / Tools

  • Android (Kotlin, Jetpack Compose)
  • Firebase (including Firebase AI Logic for on-device API calls)
  • Gemini (prompt-based reasoning for image and decision)
  • OpenAI Whisper (speech-to-text)
  • Figma (flow prototyping)

Collaboration

  • Conducted as a lab study project with Prof. Eun-Kyung Choe’s team (University of Maryland).
  • I owned the full engineering scope: mobile implementation, AI integration, and prompt design.

Outcome

A working multimodal prototype that produces a clear three-level safety verdict (“Do not use / Consult a doctor / Safe for use”) with explanations. The prototype was developed and iterated within a lab-study setting.