An AWSome AI'd

Keep it short! He's smart!

Comment

Inspiration

Modern smartphone users face several challenges: App Fragmentation: Average user has 80+ apps, creating cognitive overhead Complex Workflows: Simple tasks require multiple app switches and navigation steps Accessibility Barriers: Traditional UIs aren't optimized for voice/gesture interaction Context Loss: Apps don't share context, requiring repetitive input Learning Curves: Each app has unique interfaces and interaction patterns

What it does

Our Solution: A single AI agent that understands natural language, learns user preferences, and orchestrates actions across apps and services autonomously.

How we built it

Amazon Bedrock Agent (AgentCore) ⭐

Implements multi-step reasoning and action planning Uses action groups for each capability domain Manages conversation context and disambiguation Coordinates between input processing and action execution

Amazon Bedrock - Nova Models ⭐

Nova Micro/Lite: Intent recognition, parameter inference, decision-making Nova Vision: Image understanding, OCR, object identification Provides low-latency reasoning for real-time interactions

AWS Lambda

Orchestrator function: Manages request lifecycle Action handlers: Execute specific tasks (7 handler functions) Serverless architecture ensures scalability and cost-efficiency

Amazon API Gateway

RESTful endpoints for frontend communication Request validation and transformation CORS support for web-based access

Amazon DynamoDB

User preferences table: Stores learned behaviors Conversation context table: Maintains session state TTL-enabled for automatic cleanup

Amazon S3

Stores captured images temporarily Hosts frontend static assets Configured with CORS for direct browser access

Reasoning LLM: Amazon Nova Micro/Lite performs multi-step reasoning to: Parse complex user intents Infer missing parameters from context Plan sequential actions Handle ambiguity through confirmation flows

Autonomous Capabilities: Learns from user interactions (preference building) Makes decisions without explicit instructions Handles multi-turn conversations Proactively suggests actions based on context

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for An AWSome AI'd

Phase 2 (Post-Hackathon): Native Mobile Apps: React Native with offline capabilities More Integrations: Uber, Zomato, Amazon, Spotify Smart Home Control: Alexa device integration Proactive Suggestions: "Time to reorder milk" notifications Multi-user Support: Family accounts with shared preferences

Phase 3 (Advanced Features): Custom Voice Models: Personalized TTS/STT Gesture Control: Hand tracking for hands-free operation AR Integration: Visual overlays for object information Collaborative Agents: Multiple agents working together Enterprise Features: Calendar, email, productivity tools

Built With

Share this project:

Updates