Public-sector teams can now connect evidence stored on AWS to a conversational agent in Salesforce without placing raw media directly in the chat. A reference architecture uses Amazon Bedrock Data Automation to extract structured information from body-camera footage, surveillance video, audio, photographs and scanned records.
When a file reaches Amazon S3, an event starts an asynchronous processing job. Documents can yield text, key fields and summaries; images can produce descriptions and recognized text; and audio or video can produce transcripts and scene-level summaries. The results are stored separately and indexed with metadata in DynamoDB.
A Salesforce Agentforce user can then ask a natural-language question. The request travels through the Model Context Protocol, or MCP, to an AgentCore gateway, which authenticates it and calls an AWS Lambda service. That service locates the processed result and returns it to the agent’s context.
The pattern reduces manual review and keeps large source files outside the conversation, but extracted evidence is not automatically authoritative. Agencies must validate permissions, retention rules and model output, especially when a summary or object-detection error could affect an investigation or public decision.