Inspiration
Operators at water treatment plants and river monitoring sites are not always watching a dashboard. A missed alert about low oxygen or a pH spike can mean a delayed response. A dashboard notification is easy to miss, and a phone call is not.
What it does
AI incident response for water quality monitoring. HydroWatch detects anomalies in live sensor data (pH, temperature, dissolved oxygen), stores them, and phones the on-duty operator. On the call, the operator can ask for live readings, alert history and procedures grounded in SOP documents, and acknowledge the alert by voice.
How we built it
Built a modular backend using Spring Boot microservices integrated with Redis Streams for event-driven telemetry ingestion and Docker Compose for containerization. Sensor telemetry data (pH, dissolved oxygen, temperature, flow rate) is persisted in PostgreSQL and SQLite. For automated incident response, we integrated a custom Bolna AI voice agent via Twilio webhooks, which triggers automated phone alerts to operators when critical water parameters exceed safe thresholds.
Challenges we ran into
1). Ensuring seamless, real-time handoffs between backend alert triggers and the Bolna AI voice pipeline to prevent lag during emergency phone notifications. 2). Managing high-throughput concurrent data streams from multiple IoT nodes without blocking the main event loop. We solved this by implementing Redis Streams for asynchronous message buffering and decoupled microservice communication.
Accomplishments that we're proud of
1). Successfully engineering an end-to-end event-driven architecture that bridges IoT telemetry with real-time generative voice automation. 2). Integrated a live AI voice agent that autonomously places phone calls to operators with context-aware incident summaries during simulated water quality anomalies. 3). Maintained low latency and high system correctness across containerized microservices under simulated load.
What we learned
1). The power of event-driven design using Redis Streams for decoupling heavy sensor ingestion pipelines from notification microservices. 2). Designing robust webhook payload handlers and state management for conversational AI voice agents in emergency alert scenarios.
What's next for HydroWatch
1). Agent should be capable of routing a technician and made payment through PayPal payment Sandbox after the problem is fixed; confirmed by operator. 2). Introducing machine learning models to forecast water quality degradation trends before critical thresholds are breached. 3). Expanding alerting beyond voice calls to include automated SMS, WhatsApp, and dashboard priority dispatching.
Built With
- api
- docker
- integration
- java
- microservices
- mvc
- postgresql
- redis
- spring
- streams
