Source-linked AI summary
Large Language Model based air quality monitoring and localized alert generation
Ricardo Vieira, Luis Tavares, Kaylane Lima, Lucas De Souza, Arthur Poggy, João Lima, Vitor Pinheiro, Markus Endler
TL;DR
Poor indoor air quality can harm occupants, while existing monitoring systems often provide measurements without personalized, actionable health interpretation. EnQyMo combines IoT sensing, BLE-based localization, and literature-informed LLM agents to identify exposure risks and notify affected users. The platform demonstrated low-latency, scalable localized alert generation, but evaluation remains limited to controlled laboratory simulations.
Problem
Existing indoor-air monitoring systems are largely passive, reporting pollutant measurements without real-time, context-aware health insights for specific people at risk.
Method
EnQyMo combines IoT middleware, BLE-based presence tracking, preprocessing, and agentic LLMs that retrieve and interpret scientific and medical knowledge about pollutant-related health risks.
Results
The platform achieved average RTTs below 100 ms even at peak tested capacity, with 95% of events at 100 nodes completing under 73.67 ms.
Takeaways & Limitations
EnQyMo shifts indoor-air monitoring toward proactive, occupant-aware alerts that deliver localized warnings to employees exposed to specific health risks.
Takeaways & Limitations
Experiments used controlled laboratory simulations rather than longitudinal deployment in complex, densely occupied buildings, leaving sensor drift and BLE-obstruction effects insufficiently evaluated.
Abstract
from arXiv · showhide
Poor indoor air quality can cause up to five times more direct health problems to occupants than outdoor air. In particular, it may cause headaches, fatigue, eye/throat irritation, and long-time exposure is linked to respiratory and heart as well as some forms of cancer. Despite the importance of indoor health and well-being, most current monitoring devices and systems (usually for offices and workspaces) are passive. The Environmental Quality Monitor (EnQyMo) platform is a generic Internet of Things (IoT) middleware designed to process several sensor data related to air quality in indoor spaces and correlate this data with health exposure risks of users/workplace employees. Using Bluetooth Low Energy (BLE) beacons and a mobile IoT middleware it is able to identify the (smartphone) users exposed to these polluted air or high CO2 (carbon dioxide) levels, and generate location-specific alarms only to the users at the places with the unhealthy air conditions. At the core of EnQyMo is an agency of Large Language Models (LLMs) capable of interpreting regulatory standards and scientific literature to automatically identify critical health exposure levels.
1. Introduction
EnQyMo addresses the gap between passive indoor-air monitoring and personalized, real-time health protection. It combines literature-informed LLM inference with localized sensing and notifications for people exposed to unhealthy conditions.
- 1. Introduction: Indoor air pollution is linked to immediate symptoms including headaches, fatigue, and irritation, while prolonged exposure is associated with respiratory and heart disease and potentially some cancers.
- 1. Introduction: Most current monitoring systems provide raw pollutant measurements without translating them into actionable, person-specific health insights or real-time contextual interventions.
- 1. Introduction: EnQyMo uses an agentic LLM framework to interpret scientific literature and medical guidelines, defining dynamic hazard intervals for pollutant-related health risks.The framework is intended to correlate pollutant levels with impacts on chronic respiratory conditions.
- 1. Introduction: BLE beacons and ContextNet/Mobile Hub middleware identify people in hazardous zones and enable targeted notifications based on location and exposure duration.A preprocessing layer filters sensor noise before LLM-driven inference.
- 1. Introduction: The paper focuses on sensing, intelligent inference, and targeted notification, while autonomous HVAC or filtration control remains part of EnQyMo’s longer-term vision.
2. Related Work
Related work spans pollutant detection, prediction, conversational access, thermal comfort, and explainable diagnosis. EnQyMo positions itself as an integrative, health-oriented platform that connects semantic reasoning with occupant-aware alerts, while remaining less mature than narrower systems.
- 2. Related Work: The literature emphasizes modularity and scalability as important for adapting monitoring systems to different built environments and occupational-health contexts.
- 2. Related Work: Unlike thermal-comfort and conversational systems, EnQyMo emphasizes correlations between multimodal pollutant profiles and respiratory health conditions to generate proactive exposure alerts.
- 2. Related Work: Its explainability goal includes justifying why an alert was issued for a specific person, zone, and time, not only diagnosing IAQ degradation.
- 2. Related Work: EnQyMo adds scientific semantic reasoning to statistical pollutant-pattern detection, enabling assessment of health implications alongside concentration trajectories.
- 2. Related Work: EnQyMo combines mobile-hub sensing, specialized LLM agents, scientific-knowledge retrieval, and occupant-aware alerts within one integrated IAQ architecture.The comparison describes this combination as capabilities that are largely disconnected in existing literature.
- 2. Related Work: The related-work comparison identifies a trade-off between scope and maturity: narrower systems are more mature, whereas broader integrated proposals such as EnQyMo remain earlier-stage.
3. Previous Works and Underlying Technologies
The underlying platform combines ContextNet’s scalable backend services with smartphone-based Mobile Hubs that connect peripheral IoT devices. An N8N pipeline historizes sensor data and constructs a retrieval-augmented scientific knowledge base for AI agents.
- 3.1. ContextNet Kafka Core: ContextNet Kafka Core provides scalable IoT communication through Gateway, GroupDefiner, PoA Manager, and Mobile Temporary Disconnect microservices.The services run on cloud servers or virtual machines and support mobile edge devices called M-Hubs.
- 3.1. ContextNet Kafka Core: The CKC architecture connects mobile nodes to a Gateway through MR-UDP, while Kafka Brokers connect the Gateway and other middleware services.
- 3.2. Mobile Hub: Mobile Hubs run on conventional smartphones, discover and register peripheral IoT devices, and support remote unicast and group-cast communication.They bridge CKC internet connectivity with short-range wireless connections to sensors, beacons, and actuators.
- 3.3. AI Platform: The AI platform uses N8N to orchestrate data historization and retrieval-augmented knowledge-base construction from scientific documents.Vectorized papers and technical standards support retrieval by the AI agents.
4. Platform for Air Quality Monitoring and Alert Generation: an Overview
EnQyMo combines environmental sensing, mobile middleware, BLE-based presence tracking, and focused alerting for indoor air-quality monitoring. Its symbolic location strategy identifies affected rooms and enables notifications targeted to exposed users.
- 4.1. Air Quality Sensor Device: The IAQ device combines an ESP32 with Wi-Fi and BLE, sensors for temperature, pressure, humidity, VOCs, particulates, and CO2, plus local SD-card backup.Sensor data is periodically transmitted over Wi-Fi to the INSPIRE service while also being backed up locally.
- EnQyMo re-engineers ContextNet modules and services across mobile edge and backend components to implement monitoring, alerting, and focused notifications for affected users.The platform’s extensions support monitoring intelligence and notify only users directly affected by unhealthy environmental conditions.
- 4.3.1. Flutter-ready Mobile-Hub components: The mobile application was migrated from native Kotlin into a reusable Flutter plugin, preserving native M-Hub access while supporting Android and iOS deployment from one codebase.Flutter Platform Channels connect Dart with Kotlin through method calls for control and event channels for continuous sensor and status updates.
- 4.3.2. Symbolic Location Map: BLE beacons and smartphone scans create contextual observations that deterministic Group Definer rules map to functional rooms rather than absolute coordinates.The resulting symbolic statement identifies a user as present in a specific room, avoiding complex localization models and reducing infrastructure dependencies.
- 4.3.2. Symbolic Location Map: The symbolic location mapping generates real-time user groups and sends targeted alerts only to workers exposed to adverse environmental conditions.This localized strategy connects inferred room presence with focused notification delivery.
5. Intelligent Scientific Publication Information Retrieval Engine (INSPIRE)
INSPIRE transforms sensor readings and scientific knowledge into health-oriented alerts through specialized AI agents and retrieval-based knowledge bases. Its workflow detects abnormal measurements, correlates pollutants with diseases, and routes localized notifications to exposed users.
- Architecture: INSPIRE is the core functional module that transforms raw sensor data into actionable health alerts through specialized microservices and AI agents.Its architecture includes data acquisition, analysis, routing, persistence, publication, and disease-correlation components.
- Data processing: Sensor data are normalized, persisted, and analyzed over configurable intervals to identify readings outside health and occupational safety standards.The analyzer queries historical readings and generates an alert message when a pollutant exceeds an acceptable range.
- Localized dissemination: The completed alert is persisted and published through ContextNet so Mobile Hubs can deliver it to users located near the affected sensor.The workflow targets patients or collaborators in the sensor’s location rather than broadcasting indiscriminately.
- Disease correlation: A disease-correlation agent uses scientific documents stored in a vector database to retrieve and infer Disease × Pollutant associations for exposed people.The resulting analysis is formatted as JSON for downstream dissemination.
- Alert detection: A first AI agent uses technical standards to detect threshold violations and produces JSON messages identifying the sensor, pollutant, and exceedance level.These messages are routed to agents specialized for the relevant environmental data type.
6. Early Experiments
A laboratory proof-of-concept deployed EnQyMo end to end for sensing, inference, and targeted alert delivery. The system detected simulated pollution events, inferred associated health risks, delivered alerts to colocated smartphones, and validated photovoltaic BLE beacon proximity detection.
- Environmental sensing: The monitored device streamed real-time measurements including 998 ppm CO2, 24.3°C temperature, 66.1% relative humidity, and particulate-matter concentrations to INSPIRE.These readings were visualized on a dashboard displaying critical metrics.
- Inference pipeline: The workflow received sensor data, compared measurements with safety thresholds, triggered disease analysis, and published a structured alert to Kafka.Example inferred risks included asthma, bronchitis, and respiratory irritation.
- Targeted alert delivery: Three colocated smartphones received the same targeted JSON alert identifying PM2.5, a moderate risk level, and inferred health risks.The result confirmed ContextNet GroupDefiner and Mobile-Hub operation for alerting users in a hazardous environment.
- Beacon validation: Photovoltaic BLE beacons transmitted RSSI-based proximity signals, and varying RSSI values allowed the mobile application to distinguish distance and zone proximity.The prototype used a CN3791 solar charge controller, ESP32 microcontroller, and 18650 battery.
- Limitations: The evaluation was an early proof-of-concept in a controlled laboratory with simulated pollution events, so it did not assess long-term deployment challenges such as sensor drift or BLE obstruction effects.Future work targets multi-room validation, scalability, hardware durability, and practical notification utility.
7. Simulation based experiments
Simulation-based stress tests evaluated localized alert distribution with up to 100 concurrent target nodes using static Java-emulated clients. Average RTT remained below 100 ms, while latency distributions increased modestly and stayed predictable at peak load.
- Experimental design: The stress tests used Java-emulated mobile nodes and static beacon assignments to isolate ContextNet gateway and alert-distribution performance from wireless, battery, and mobility variability.Static mode also avoided fluctuating group sizes and acknowledgment counts caused by dynamic beacon hopping.
- Experimental design: The simulation architecture sent updated contexts to the Group Definer while a separate node injected a mocked INSPIRE alert into the Processing Node.The runner staggered node initialization by 100 milliseconds and assigned unique UUIDs.
- Latency results: Average RTTs remained well below 100 ms while alert distribution scaled from 5 to 100 concurrent target nodes, indicating efficient scalability.The evaluation measured gateway dispatch time plus acknowledgments from target nodes.
- Latency results: At 50 simultaneous acknowledgments, mean RTT increased only marginally to 39.24 ms after low-density values of approximately 29–33 ms.A single 90.20 ms result at five nodes was attributed as likely JVM warmup overhead.
- Distribution analysis: Median RTT rose from 24.00 ms at 5 nodes to 46.44 ms at 100 nodes, while 95% of events at 100 nodes completed below 73.67 ms.Maximum spikes included 90.20 ms at five nodes and 84.51 ms at peak load.
- Conclusion: The authors conclude that the architecture supports near-real-time localized alerts for large indoor populations without observed network or processing bottlenecks.This conclusion is limited to the simulated static-mode evaluation.
8. Conclusion
EnQyMo demonstrates a general framework combining AI agency, LLMs, scalable IoT middleware, and zone-based BLE localization for proactive occupational-health alerts. The current implementation validates the core architecture, while broader environmental sensing, edge processing, and long-term beacon autonomy remain future work.
- 8. Conclusion: The platform integrates low-cost IoT hardware, ContextNet, and Mobile-Hub middleware into a pipeline that contextualizes environmental variables as actionable health insights.
- 8. Conclusion: EnQyMo combines AI agency, LLMs, scalable IoT middleware, and occupant-aware alerting to shift occupational-health monitoring from passive data logging toward proactive warnings.The framework interprets sensor readings using health knowledge obtained from scientific and medical literature.
- 8. Conclusion: Solar-powered BLE beacons provide zone-based localization without costly positioning infrastructure or routine battery replacement, while supporting cross-platform deployment through a Flutter plugin.
- 8.1. Future Work: Future work will expand monitoring beyond CO2 and particulate matter, use smartphones as mobile sensor gateways, and activate edge-based complex-event processing for local anomaly detection.
- 8.1. Future Work: Long-term field studies are still needed to validate photovoltaic beacon energy-harvesting efficiency under varying indoor lighting conditions.