AquaQuest AI

Empowering communities to monitor and protect urban water quality through AI vision, real-time telemetry, and gamified citizen science.

Comment

IEEE OneAquaHealth Hackathon 2026 Horizon Europe Platform Backend Standards

Inspiration

Urban freshwater ecosystems—rivers, streams, canals, and urban wetlands—are vital indicators of ecological stability and public health. Under the One Health paradigm, human health, animal welfare, and environmental integrity are deeply intertwined. However, community-led water monitoring faces three persistent barriers:

  1. The Usability & Jargon Barrier: Ecological metrics (e.g., Nephelometric Turbidity Units, dissolved oxygen saturation, benthic macroinvertebrate indices) rely on complex scientific terminology that alienates non-specialist citizens.
  2. Low Retention & Single-Use Churn: Over 80% of contributors on existing citizen science platforms log only a single observation before abandoning the platform.
  3. Data Silos & Lack of Verification: Crowdsourced environmental data is rarely standardized or verified, preventing its integration into official healthcare and public policy frameworks.

Inspired by the IEEE OneAquaHealth Global Hackathon 2026 and the 5 European Pilot Cities (Coimbra, Gent, Benevento, Oslo, and Toulouse), we created AquaQuest AI—a platform that transforms smartphones into intelligent environmental diagnostics tools, bridging citizen science, computer vision, and digital health standards.

What it does

AquaQuest AI is a native Android application paired with a dual-stage AI backend that empowers citizens to monitor, verify, and safeguard local water health:

  • 📷 Smart CameraX Viewfinder: Guides users in real-time with surface detection, framing guides, and Google ML Kit on-device filtering to discard non-water frames before analysis.
  • 🔬 Dual-Layer Vision & AI Spectrometry: Analyzes captured water photos using Python OpenCV spectrometry (color histograms, turbidity proxies, microplastic surface risk, and oil sheen indicators) combined with multimodal LLM reasoning to produce plain-English diagnostic advisories.
  • 🌐 USGS Live Telemetry & OpenStreetMap Hub: Integrates real-time USGS stream gauge data (discharge, stage height, pH, dissolved oxygen, temperature) with an interactive vector map of local river networks.
  • 🎮 Geofenced Verification Quests: When an anomaly (risk score > 6/10) is logged, automated proximity notifications (<2 km) invite nearby citizens to cross-verify the observation, rewarding participants with XP, streak bonuses, and unlockable badges (Stream Scout to Master HydroGuardian).
  • 🏥 Standardized Digital Health Exporter: Generates 1-tap HL7 FHIR R4 Observation JSON (with LOINC codes 21612-7, 2748-2, 21611-9) and ISO 19115 GeoJSON payloads, bridging citizen observations directly into clinical and public health databases.

How we built it

AquaQuest AI was built using a robust modern tech stack tailored for mobile performance, computer vision, and health data interoperability:

  • Mobile Front-End: Built natively in Kotlin using Jetpack Compose, following modern MVVM architecture, Material 3 design system, and Room DB for offline persistence.
  • On-Device Computer Vision: Integrated Google ML Kit within the CameraX workflow to perform zero-latency frame validation before remote processing.
  • Python OpenCV Spectrometry Server: Developed a Flask backend incorporating OpenCV pipeline algorithms (HSV color segmentation, Canny edge density for turbidity estimation, and specular highlight reflection analysis for oil sheens).
  • Multimodal AI Reasoning: Connected the vision pipeline with multimodal LLM APIs (via Groq/Qwen) to ground qualitative computer vision data into plain-English advisories for human contact, pet safety, and aquatic wildlife.
  • Geospatial & Hydrographic Data: Integrated the USGS Water Data REST API for real-time sensor streams and OpenStreetMap (Overpass API) for vector layer rendering of urban stream corridors.
  • Interoperability Engines: Engine modules (FhirWaterObservationExporter, OneHealthExporter) map environmental telemetry into HL7 FHIR R4 Observation structures and ISO 19115 compliant spatial metadata.

System Architecture

graph TD
    subgraph Client ["📱 Android Native App (Kotlin / Jetpack Compose)"]
        UI["UI Layer (Compose & Material 3)"]
        VM["ViewModel & State Management"]
        Cam["CameraX + ML Kit (On-Device Water Filter)"]
        LocalDB[("Room Local DB (Offline Cache)")]
        FhirEng["FHIR R4 & ISO 19115 Export Engine"]
    end

    subgraph Backend ["🐍 Python AI & Spectrometry Server (Flask)"]
        CV["OpenCV Pipeline (Turbidity, HSV, Oil Sheen)"]
        LLM["Multimodal AI Reasoning Engine"]
    end

    subgraph External ["🌐 External Data & Health Standards"]
        USGS["USGS Water Data REST API"]
        OSM["OpenStreetMap Overpass API"]
        HealthSys["HL7 FHIR Clinical & Public Health Systems"]
    end

    Cam -->|"Validated Water Frames"| CV
    CV -->|"CV Spectrometry Metrics"| LLM
    LLM -->|"Plain-English Diagnostics & Advisories"| VM
    USGS -->|"Live Telemetry (pH, Temp, DO)"| VM
    OSM -->|"Vector River Corridor Mapping"| UI
    VM <--> LocalDB
    UI --> FhirEng
    FhirEng -->|"1-Tap LOINC FHIR R4 Observation JSON"| HealthSys

System Data Flow

+-----------------------------------------------------------------------------------+
|                           📱 AquaQuest AI Mobile App                             |
|  +---------------------+   +---------------------+   +-------------------------+  |
|  | Jetpack Compose UI  |---|  ViewModel & State  |---| Room DB Local Storage   |  |
|  +---------------------+   +---------------------+   +-------------------------+  |
|             |                         |                           |               |
|  +---------------------+              |              +-------------------------+  |
|  | CameraX + ML Kit    |              |              |  HL7 FHIR R4 Exporter   |  |
|  | (On-Device Filter)  |              |              |  (LOINC Standardized)   |  |
|  +---------------------+              |              +-------------------------+  |
+-------------|-------------------------|---------------------------|---------------+
              | (Image Payload)         | (Telemetry Request)       | (FHIR JSON)
              v                         v                           v
+-----------------------------+   +-------------------+   +-------------------------+
| 🐍 Python Spectrometry      |   | 🌐 USGS Telemetry |   | 🏥 Digital Health &     |
|    Server (OpenCV & Flask)  |   |    & OSM Services |   |    Public Health Systems|
+-----------------------------+   +-------------------+   +-------------------------+
              |                                
              v                                
+-----------------------------+                
| 🤖 Multimodal AI Engine     |                
|    (LLM Diagnostic Reasoning)|                
+-----------------------------+

Challenges we ran into

  1. Translating Technical Metrics into Plain Language: Environmental sensors output raw numerical data (e.g., specific conductance in µS/cm or turbidity in NTUs). Crafting context-aware rules to translate these values into clear recommendations for swimming, pets, or wildlife without oversimplifying required extensive domain research.
  2. Computer Vision in Uncontrolled Outdoor Environments: Sunlight glare, water ripples, and varying shadow conditions frequently interfere with OpenCV edge and color metrics. We addressed this by implementing adaptive thresholding, specular reflection masking, and pairing computer vision metrics with LLM multimodal reasoning.
  3. Bridging Environmental Citizen Science with Clinical Standards: Mappings between ecological observations and HL7 FHIR R4 healthcare schemas are uncommon. We spent significant effort standardizing LOINC code mappings (21612-7 turbidity, 2748-2 pH, 21611-9 temperature) to ensure exported FHIR payloads pass formal validation.
  4. Gamification without Data Spoofing: Designing a quest engine that incentivizes frequent observations while preventing fake reports required developing spatial-temporal consensus algorithms based on geofenced peer verification within a 2 km radius.

Accomplishments that we're proud of

  • Comprehensive 7-Track Fusion: Successfully unified all seven hackathon focus tracks—Citizen Science UX, Data-to-Insight, AI Assessment, Storytelling, Gamification, Resilience Informatics, and Digital Health Standards—into a single application.
  • Clinical Health Interoperability: Created an automated 1-tap HL7 FHIR R4 exporter, enabling citizen-sourced water observations to be ingested by public health systems.
  • Real-Time USGS Telemetry Integration: Connected live stream gauge data across thousands of monitoring stations with custom One Health risk interpretation layers.
  • Tailored European Pilot Baselines: Configured baseline water parameters for 5 European cities (Coimbra, Gent, Benevento, Oslo, Toulouse), highlighting local hydrological challenges like thermal drying, cyanobacteria risk, and urban runoff.

What we learned

  • The Power of One Health Interoperability: Linking environmental indicators directly to human and animal health standards (HL7 FHIR) reveals how public health systems can leverage citizen science for early disease vector warnings.
  • Hybrid Vision Systems Excel: Combining fast on-device screening (Google ML Kit) with backend CV spectrometry (OpenCV) and cloud LLM reasoning yields higher diagnostic accuracy while minimizing network payload size.
  • Proximity Gamification Drives Consensus: Geofenced quests not only boost user engagement but also serve as an effective data quality control layer through crowdsourced verification.

What's next for AquaQuest AI

  • On-Device Edge Processing (TFLite / NCNN): Porting OpenCV spectrometry directly to an on-device TensorFlow Lite model so citizens can analyze water clarity in offline river basins without cellular connectivity.
  • IoT Sensor Integration: Supporting low-cost Bluetooth LE water probes (pH, TDS, temperature) to auto-sync physical measurement data with photo captures.
  • Municipal Early-Warning Dashboard: Building a web portal for urban planners and public health officials to view real-time anomaly clusters, predictive bio-risk trends, and deploy targeted interventions.
  • Global River Basin Expansion: Expanding pre-configured river basin profiles beyond European pilot sites to vulnerable watersheds globally.

Built With

Share this project:

Updates

Submission history