AWS Builder Center

Builder Center

Building better_AQI: Hyper-Local Pollution Intelligence with AWS Strands Agents, SAM CLI, and LocalStack

How we built a serverless environmental decision platform that turns abstract air quality numbers into life-saving commute, school, and indoor action protocols.



Building better_AQI: Hyper-Local Pollution Intelligence with AWS Strands Agents, SAM CLI, and LocalStack

Every winter, millions of citizens across Indian metropolitan hubs like Delhi NCR wake up under a dense, toxic smog blanket. While standard weather apps report a single, abstract Air Quality Index (AQI) number recorded at a distant monitoring tower kilometers away, that number fails to answer critical everyday questions:
  • Which commute route exposes my lungs to the least toxic particulate matter?
  • Is it safe for 800 school students to conduct outdoor morning assembly today?
  • When will seasonal stubble smoke plumes reach our neighborhood?
  • How many minutes must we run our classroom HEPA purifier to reach medical-grade safe air?
To bridge this gap between raw environmental sensors, satellite telemetry, and human safety, we built better_AQI for Track 01 (Air) of the WeMakeDevs & AWS Environmental Hacks 2026.
In this post, we'll dive into the architecture of better_AQI, how we leveraged the AWS Strands Agents SDK powered by Amazon Bedrock, and how we used the AWS SAM CLI and LocalStack to achieve a completely reproducible serverless deployment.


Architecture Overview: Serverless by Design

better_AQI is built from the ground up following AWS serverless best practices. Rather than running persistent, costly EC2 instances, our platform scales to zero during off-peak hours and instantly absorbs traffic spikes during peak morning commute and evening pollution alerts.
Architecture Flow: [ React 19 Frontend (Vite) ] | v [ Amazon API Gateway (HTTP API) ] | v [ AWS Lambda (Python 3.12 + Mangum) ] <--- [ AWS EventBridge (Hourly Cron) ] | | | v v v [ Open-Meteo & [ Amazon [ Amazon SNS NASA FIRMS ] DynamoDB ] Spike Alerts ] (Subscriptions)

Core AWS Services Used:

  1. AWS Lambda (BetterAqiApiFunction): Executes our Python 3.12 FastAPI backend through the Mangum ASGI adapter, maintaining sub-100ms response times.
  1. Amazon API Gateway (BetterAqiHttpApi): Manages TLS termination, global CORS, and low-latency route dispatch.
  1. Amazon DynamoDB (better_aqi_subscriptions): NoSQL document store persisting user coordinates, alert thresholds, and notification preferences.
  1. Amazon SNS (better_aqi_spike_alerts): Manages pub/sub fan-out for real-time SMS and email alerts during severe smog spikes.
  1. AWS EventBridge (HourlyAirQualityCheckSchedule): Evaluates live environmental sensors against subscriber alert thresholds every hour.
  1. Amazon Bedrock: Foundation models powering our agentic environmental reasoning engine.


Orchestrating Environmental Intelligence with AWS Strands Agents

A key innovation in better_AQI is the AI Air Advisor, built using the open-source AWS Strands Agents SDK (strands-agents) and Amazon Bedrock.
Traditional chatbots often hallucinate safety metrics. With Strands Agents, the model is equipped with typed Python tools connected directly to our scientific calculation engines:
  • Tool 1: check_live_air_quality(city) -> Fetches live multi-pollutant levels (PM2.5, PM10, AQI) calibrated to Indian CPCB NAAQS breakpoints.
  • Tool 2: evaluate_school_safety(pm25) -> Evaluates assembly, recess, and mid-day ventilation safety for schools.
  • Tool 3: calculate_commute_inhalation(route, mode) -> Calculates inhaled toxic PM2.5 dosage and Berkeley Earth cigarette equivalents.
When a school principal asks, "Can we hold sports day outside today in Noida?", the Strands Agent executes check_live_air_quality, verifies the PM2.5 level against CPCB thresholds, calls evaluate_school_safety, and delivers an authoritative, actionable recommendation rather than generic conversational filler.


Infrastructure as Code with AWS SAM CLI and LocalStack

To ensure our cloud infrastructure is 100% reproducible and verifiable, we defined our entire stack in a concise AWS SAM template (template.yaml):
  • Serverless Function: Defined using Python 3.12 runtime with HTTP API Gateway proxy integration.
  • Database: Modeled as a pay-per-request Amazon DynamoDB table with a single partition key.
  • Notifications: Declared an Amazon SNS topic for instant spike alert fan-out.
  • Scheduling: Created an AWS EventBridge scheduled rule running hourly evaluations.

Zero-Cost Cloud Emulation with LocalStack

During early development, we avoided cloud deployment friction by using LocalStack in Docker:  docker compose -f docker-compose.localstack.yml up -d 
This allowed our automated pytest suite to test DynamoDB subscriptions and SNS notifications locally with zero cloud costs and zero credential dependencies.


Scientific Rigor: Moving Beyond Abstract Scores

better_AQI integrates peer-reviewed environmental models into the user experience:
  • Cleanest Commute Engine: Calculates minute-by-minute respiratory uptake across transit modes (Bicycle, Two-Wheeler, Car with AC Cabin Filter, Walking) and displays inhaled toxic dosage in Berkeley Earth cigarette equivalents (22 ug PM2.5 / 24hr ≈ 1 cigarette).
  • NASA Stubble Burning Plume Tracking: Combines active NASA satellite thermal hotspot coordinates across Punjab and Haryana with live boundary-layer wind vectors to track downwind smoke corridors.
  • Classroom CADR Countdown: Models room volume and building envelope leakage to provide a live countdown timer until a HEPA purifier achieves safe indoor air (< 30 ug/m3).


Key Learnings & Developer Takeaways

  1. Agent Tool-Calling vs. Prompt Stuffing: Giving an LLM dedicated, typed Python tools via AWS Strands Agents produces orders of magnitude more reliable answers than prompting a model with raw context blobs.
  1. SAM Simplifies Serverless: Declaring API Gateway routes, Lambda triggers, DynamoDB tables, and EventBridge cron rules in under 120 lines of SAM YAML saved hours of AWS console clicking.
  1. Cloud Parity Accelerates Velocity: Testing serverless storage and pub/sub pipelines locally via LocalStack enabled us to write comprehensive unit and integration tests before touching the cloud.


Get Involved

  • Built for: WeMakeDevs & AWS Environmental Hacks 2026 (Track 01: Air)
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article