AeroSentinel

Inspiration## Inspiration

Traditional air quality systems only act as historical dashboards, showing pollution levels after the toxic air has already been inhaled. We wanted to build a proactive system that gives municipal authorities a clear lead time to dynamically protect citizens before a pollution event occurs.

What it does

AeroSentinel shifts urban air quality management into an automated Predict-Optimize-Verify-Act lifecycle. It ingests real-time CPCB API data, forecasts localized AQI 24 hours in advance, and automatically routes verified safety response plans to minimize toxic exposure in high-risk zones.

How we built it

• Predictive ML: Built using an XGBoost Regressor for dynamic 24-hour forecasting, integrated with a SHAP framework for explainable transparency.
• Multi-Agent Orchestration: Deployed a dual hierarchical system via CrewAI consisting of a Routing Agent and a Policy Simulation Agent.
• Safety Layer: Implemented a hard-coded Python Deterministic Validation Gate to eliminate AI hallucinations.
• Frontend: Interactive Streamlit framework utilizing Folium maps for live dynamic spatial visualization.

Challenges we ran into

Handling real-time API dropouts and structural gaps in sparse data streams. We resolved this by building robust automated data ingestion wrappers featuring fallback imputation modules.

Accomplishments that we're proud of

Successfully bridging the gap between raw predictive machine learning models and deterministic, actionable safety frameworks that can run autonomously without administrative delays.

What we learned

Deepened our understanding of agentic workflows, multi-agent resource optimization logic, and how to safely bind generative LLM planning with strict, hard-coded execution guardrails.

What's next for AeroSentinel

Integrating real-time automated traffic routing API webhooks to dynamically divert vehicle clusters away from school and hospital zones during peak pollution windows.

What it does## What it does

• Predictive Environmental Alerts: Forecasts localized Air Quality Index (AQI) surges 24 hours in advance by analyzing real-time pollution datasets.
• Proactive Crisis Control: Automatically triggers dynamic emergency routing protocols (like dust suppression or traffic diversion) instead of just tracking historical data.
• Interactive Operations Portal: Provides municipal authorities with an intuitive frontend console featuring real-time spatial heatmaps and live sensor health updates.

How we built it

• Robust Ingestion Architecture: Designed live Request API wrappers targeting CPCB, OpenAQ, and OpenWeather with integrated rolling-mean imputation for sensor data gaps.
• Explainable Intelligence Core: Integrated an XGBoost Regressor pipeline tightly coupled with the SHAP validation framework to ensure transparent feature attribution for every prediction.
• Orchestrated Multi-Agent System: Configured a hierarchical reasoning layer utilizing CrewAI and LangChain, where a Routing Agent flags high-risk grid zones and a Policy Agent runs resource optimization simulations.
• Deterministic Execution Gateway: Implemented a strict Python-based validation logic block to cross-examine generative agent outputs against strict hardware safety limits before rendering onto a Streamlit dashboard.

Challenges we ran into

• Managing Sensor Disconnections: Dealt with intermittent data drops and missing fields from real-time environmental API endpoints, which we solved by engineering an automatic backfill data-cleaning pipeline.
• Optimizing Agentic Latency: Faced high execution delays when processing multi-agent conversational reasoning loops, which we successfully minimized by applying strict JSON formatting guidelines and optimizing prompt constraints.

Accomplishments that we're proud of

• Adaptive Governance Logic: Successfully moved beyond rigid, hard-coded rule configurations to create a fluid, context-aware AI architecture that adjusts responses based on changing real-world metrics.
• Explainable AI Integration: Maintained absolute transparency in public infrastructure systems by integrating SHAP plots, allowing operators to trace exactly why a localized alert was triggered.

What we learned

• Geospatial Data Wrangling: Mastered the structural complexities of managing multi-source live sensor grids and transforming them into reactive UI components.
• Balancing Hybrid Architectures: Discovered how to safely bridge fluid generative LLM task planning with deterministic structural guardrails to prevent harmful operational hallucinations.

What's next for AeroSentinel

• Scale Geographic Coverage: Expand the core infrastructure matrix file (zones.json) to map multiple dense industrial zones and neighboring urban communities globally.
• Direct Smart Infrastructure Integration: Establish direct API webhook loops targeting automated municipal hardware—such as variable traffic warning signals and industrial air-scrubber switches—to achieve an autonomous, self-healing environment.

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Untitled

Built With

• api
• crewai
• folium
• machine-learning
• multi-agent-system
• python
• shap
• streamlit
• xgboost