AWS launches agentic AI movie assistant using Nova Sonic 2.0 and Bedrock AgentCore

AWS Builds Agentic AI Movie Assistant With Nova Sonic 2.0

Amazon Web Services detailed a production-ready agentic AI system for streaming services that combines real-time voice interaction with contextual movie recommendations and scene analysis, addressing core limitations of traditional recommendation algorithms.

The Problem With Traditional ML Recommendations

Conventional collaborative and content-based filtering systems lack contextual understanding. After watching The Shawshank Redemption, a traditional system suggests more prison dramas—missing that the user might want something lighter to unwind. This context-dependent gap (time of day, mood, social setting) remains unaddressed by pattern-recognition-only approaches.

Architecture and Core Components

AWS deployed the system using:

Nova Sonic 2.0 — Amazon's latest speech-to-speech model handling real-time, bidirectional voice conversations with low latency. The model natively supports text and streaming speech inputs, with controllable personality via system prompts for on-brand responses.

Bedrock AgentCore — Orchestrates tool invocation, context management, and response curation. The system uses the Model Context Protocol (MCP) to expose AWS Lambda functions as MCP-compatible tools.

Infrastructure Stack:

• AWS Fargate containers managing session orchestration via WebSocket connections
• JWT token validation for security
• Bidirectional Smithy streaming RPC protocol for model communication
• OpenSearch + S3 Vector for semantic search and storage
• Amazon Bedrock Data Automation for video processing

Two Core Use Cases

1. Mood-Aware Recommendations: Users describe their mental state ("something fun after a long day") rather than browsing history alone. Lambda functions retrieve user affinity profiles from DynamoDB, perform hybrid semantic search across 500+ sample movies in OpenSearch, and return contextually matched recommendations.

2. Real-Time Scene Analysis: While watching, users ask questions like "who is that actor?" or "summarize what just happened?" The system:

• Uses Amazon Bedrock Data Automation to extract chapter summaries, transcriptions, timecodes, and audio segments from video
• Applies Amazon Rekognition's celebrity recognition to identify actors
• Matches user queries to movie scripts using semantic embeddings
• Returns instant contextual answers via voice

Technical Workflow

User voice commands flow through WebSocket → Nova Sonic 2.0 → Bedrock AgentCore Gateway → Lambda functions → OpenSearch/S3 Vector → Results back to Nova Sonic for voice response → Streamed to client. Complex background tasks execute asynchronously while maintaining conversational fluidity.

What This Means

AWS is positioning agentic AI as the next generation of streaming recommendation systems, moving beyond static filtering toward dynamic, conversational discovery. The combination of real-time speech, tool orchestration via AgentCore, and semantic search creates a verifiable competitive advantage over traditional ML approaches. The architecture demonstrates how production-grade agentic systems can integrate multiple AWS services for complex workflows.

For streaming platforms, this represents a pathway to differentiation through conversational interfaces. For AWS, it showcases Bedrock AgentCore and Nova Sonic 2.0 maturity for enterprise use cases. The public GitHub repository suggests AWS intends this as a reference architecture for customers building similar systems.