AllSenses — an AI guardian that protects life in real time.

Built with Kiro and Claude 3.5 Sonnet, AllSenses is an AI guardian that detects danger from audio, motion, and GPS—and sends instant SMS + live tracking when you can’t call for help.

Comment

🕯️ AllSenses — An AI Guardian That Protects Life in Real Time

⭐ Inspiration

AllSenses was inspired by one terrifyingly real question:

What happens if you’re in danger… and you can’t call for help?

Rideshares, late-night walks, isolated workplaces—many situations turn dangerous exactly when reaching your phone would escalate the threat. Most emergency systems depend on your ability to dial, speak, or press a button.

But the scariest emergencies are the silent ones.

The Kiroween theme pushed this idea further:

What if we could resurrect personal safety, stitch together multi-modal sensing and AI like a Franken-agent, and create an autonomous guardian that never sleeps?

That became the mission behind AllSenses AI Guardian—an always-awake, privacy-preserving protective agent that analyzes your environment and triggers help even when you can’t.

🧩 What it does

AllSenses is an autonomous AI safety agent that:

  • Monitors for distress through audio, motion, and geolocation
  • Uses Claude 3.5 Sonnet (via AWS Bedrock) to interpret context, intent, and threat severity
  • Detects multi-factor risks in real time

  • Automatically sends emergency alerts with:

    • Live GPS tracking
    • Contextual incident metadata
    • Inferred threat indicators (e.g., distress language, abrupt noise, location anomaly)

  • Routes SMS intelligently worldwide with a hybrid engine:

    • AWS End User Messaging (EUM) for US/Canada (10DLC compliant)
    • AWS SNS for international delivery (LATAM, EU, etc.)

  • Deletes sensitive data automatically after 24 hours

  • Runs silently in the background—hands-free and globally accessible

In essence:

It speaks when you can’t. It acts when you’re unable.

🛠️ How we built it

🧠 Architecture Overview

A serverless, event-driven AI agent built entirely on AWS and orchestrated via Kiro + Claude 3.5 Sonnet:

PWA (Frontend) ↓ API Gateway (REST) ↓ Lambda (Python AI & orchestration) ↓ DynamoDB (Incidents, GPS, TTL)  ├─ EUM (US SMS)  └─ SNS (International SMS)

And, woven through the core:

AWS Bedrock — Claude 3.5 Sonnet for emergency intent detection, threat analysis, and reasoning.

🤖 AI-Powered Emergency Brain (Claude 3.5 Sonnet)

AllSenses sends contextual signals (transcripts, motion patterns, incident history) to Claude 3.5 Sonnet via Bedrock, which helps to:

  • Interpret natural language (e.g., “help”, “stop”, “they’re following me”)
  • Distinguish real emergencies vs noise
  • Assess threat severity from multi-modal cues
  • Suggest whether to trigger alerts or keep monitoring

This lets AllSenses behave like an always-awake guardian, not a simple keyword trigger.

🔍 Multi-Modal Threat Detection

Threat scoring combines multiple indicators into a single decision function:

$$ \text{Threat Score} = \sum_{i=1}^{n} w_i \cdot f_i(x) $$

An alert is triggered when:

$$ \text{Threat Score} \geq 0.7 $$

Signals include:

  • Audio distress words
  • Abrupt audio spikes
  • Sudden movement / motion anomalies
  • Location anomalies (e.g., unexpected detours, sudden stops in unsafe areas)

Claude 3.5 Sonnet helps reason over these signals and reduce false alarms.

📡 Hybrid SMS Routing Engine

def send_sms(phone, message):
    if phone.startswith("+1"):
        # US / Canada — 10DLC compliant
        return send_via_eum(phone, message)
    else:
        # International (LATAM, EU, etc.)
        return send_via_sns(phone, message)

This hybrid design keeps US messaging fully compliant while preserving global reach.

📍 Live GPS Tracking

Location updates are stored with a 24h TTL for automatic cleanup:

ttl = int(time.time()) + 86400  # 24 hours

Emergency contacts receive a CloudFront link that shows the victim’s live-updating position on a map, so responders can track movement in real time instead of relying on a static pin.

🔐 Privacy & Security

AllSenses was designed as privacy-first, not surveillance-first:

  • AES-256 encryption for all PII in DynamoDB
  • TLS 1.3 for all communications
  • TTL-based auto-deletion after 24 hours
  • Immutable consent logs for auditability
  • Strict IAM least-privilege policies
  • Minimal data collection—only what’s needed to protect the user

👹 Challenges we ran into

🩸 1. The SMS Compliance Monster

US SMS requires 10DLC registration and has strict originator rules. EUM refused international numbers even when configuration looked correct.

  • Root cause: US-only originators
  • Solution: Hybrid routing across EUM + SNS, dynamically chosen by country code

🦇 2. The CORS Poltergeist

CORS errors persisted like a haunting—no matter what headers were added.

  • Root cause: Both Lambda and API Gateway were injecting CORS headers → duplicates
  • Fix: Let API Gateway own CORS entirely and strip CORS logic from Lambda

🪦 3. Cached Responses From Beyond the Grave

Old Lambda responses kept “resurrecting” due to CloudFront and browser caching.

Fixes included:

  • Cache-busting query parameters
  • CloudFront invalidations after critical updates
  • Incognito-mode development workflow

🧟 4. Real-Time GPS Without WebSocket Overkill

Full WebSockets felt like over-engineering for a hackathon; naive polling was too slow and costly.

  • Solution: DynamoDB with short TTL and periodic updates (~10 seconds), creating a lightweight “undead loop” that feels live but remains simple and cheap

🕷️ 5. Global Reliability & International Carriers

Testing in LATAM exposed hidden carrier restrictions and regional quirks.

  • Outcome: SNS became essential for international reliability, while EUM handled US 10DLC compliance

🏆 Accomplishments that we’re proud of

  • Built an autonomous AI guardian capable of detecting distress and acting instantly
  • Achieved < 10s end-to-end emergency response (from detection to SMS delivered)
  • Designed a privacy-first safety system with AES-256 + auto deletion
  • Delivered global SMS support across US + LATAM via hybrid routing
  • Implemented live GPS streaming with low latency and auto cleanup
  • Created a serverless skeleton architecture that scales to millions of users
  • Integrated Claude 3.5 Sonnet via AWS Bedrock as the “reasoning brain” behind decisions
  • Successfully resurrected the broken concept of “emergency systems that require interaction”

📚 What we learned

Technical Lessons

  • Serverless is ideal for real-time, bursty safety workloads
  • International SMS is far more complex than domestic (compliance, originators, carriers)
  • Multi-modal detection plus Claude-powered reasoning dramatically reduces false alarms
  • CORS must be centralized—or it turns into a haunted maze of headers
  • Caching is a trickster: always validate with explicit invalidations and incognito tests

Product Lessons

  • Safety must be simple, quiet, and nearly invisible
  • False positives destroy trust; multi-factor analysis and AI reasoning fix this
  • Users care deeply about privacy and control, not just features
  • In emergencies, every second counts—UX and backend latency both matter

Mathematical Insight

Raw false positive rate:

$$ P(\text{Alert} \mid \text{Safe}) = 0.008 $$

With weighted multi-factor scoring and AI reasoning, the operational false positive rate drops to:

$$ \text{Operational FPR} < 1% $$

This validated the threat-scoring design and the need for multi-signal correlation.

🤖 AI Co-Developer — Kiro + Claude 3.5 Sonnet

AllSenses wasn’t just built with AI—it was built through AI:

  • Kiro IDE orchestrated the development flow

  • Claude 3.5 Sonnet via AWS Bedrock translated natural-language specs into:

    • Lambda handlers
    • CloudFormation templates
    • SMS routing logic
    • Threat-scoring algorithms

This AI-assisted pipeline delivered in ~10 days what would traditionally take ~8 weeks, without sacrificing robustness or privacy.

AI didn’t replace the developer—it amplified the developer’s reach.

🚀 What’s next for AllSenses — an AI guardian that protects life in real time

Phase 2 — Advanced AI

  • Voice stress and emotion detection
  • Behavioral modeling to learn a user’s “normal” patterns
  • Predictive alerts before a situation escalates

Phase 3 — Wearable Integration

  • Apple Watch / Android Watch support
  • Heart-rate + stress biometrics
  • Fall detection and “no-movement” alerts

Phase 4 — Community & Network Features

  • Crowd-sourced safe zones and risk hotspots
  • Buddy system for mutual monitoring
  • Decentralized emergency mesh between nearby users

Phase 5 — Enterprise / Workforce Safety

  • Delivery & field-worker safety monitoring
  • OSHA-ready incident logs and analytics
  • Multi-agent safety orchestration for fleets and teams

AllSenses will evolve into a constellation of safety agents, each quietly collaborating to protect people in physical spaces—autonomously, globally, and respectfully of their privacy.

Related Videos:

🔎 Inspiration Behind the Project

The idea for AllSensesAI Guardian was shaped in part by powerful journalism from The New York Times that exposed a critical problem: many victims of violence, especially inside rideshare vehicles, are unable to call for help when danger escalates.

The reporting made something very clear: When a person is threatened into silence, restrained, or unable to reach their phone, traditional emergency tools simply fail.

This raised the defining question that drives AllSensesAI: “Who protects you when you can’t ask for help?”

AllSensesAI’s mission is to close that gap — using AI to detect silent emergencies automatically and notify someone who can intervene, in under ten seconds and without requiring any action from the user.

https://www.nytimes.com/video/business/100000010323329/ubers-sexual-assault-problem.html?smid=url-share

🎥 Full System Demonstration

A complete visual walkthrough of AllSensesAI Guardian — including the user journey, serverless AWS architecture, Claude Sonnet threat detection pipeline, DynamoDB incident handling, and hybrid SMS routing — is available in the official demonstration video:

➡️ Project Demo (YouTube): https://youtu.be/KmcC2YcngEg?si=dMQu3CI36joNgG3R

This video illustrates how all core components interact in real time, showing the 10-second emergency detection sequence, automatic alert dispatch, and live GPS tracking link generation. It is the best way to understand how the system operates as a unified safety solution.

Built With

  • aes-256
  • aiagent
  • amazon
  • amazon-bedrock
  • amazon-web-services
  • api
  • apigateway
  • claude-3-5-sonnet
  • cloudformation
  • cloudfront
  • cloudwatch
  • css
  • detection
  • dynamodb
  • eum
  • event-driven
  • geo-location
  • gps
  • html5
  • iam
  • important-for-the-jury)**-`kiro`-`kiro-ide`-`amazon-bedrock`-`claude-3-5-sonnet`-`anthropic`-`ai-assisted-development`-`conversational-development`-if-you-want
  • javascript-(es6+)
  • jwt
  • jwt-based
  • kiro
  • kms
  • lambda
  • motion
  • multi-factor
  • pwa
  • python
  • s3
  • serverless
  • sms
  • sns
  • threat
  • tls
Share this project:

Updates