Technical Architecture

Frontend: React (Next.js) and Tailwind CSS, live transcription viewer, suggested responses panel, threat alerts + sentiment meter, chatbox for whisper-to-whisper queries. Backend: All data is encypted and stored via AmazonS3 and DyamoDB.

Project Description

WhisperMind Secure is a real-time, GenAI-powered call assistant designed to support customer service agents in high-risk or regulated environments like finance, healthcare, and cloud platforms. It transcribes live conversations, detects emotional tone, suggests empathetic responses, and flags potential security incidents or sensitive data disclosures—all in real time.

Implementation Details

The application uses Amazon Transcribe to stream speech-to-text from live calls, which is then routed through AWS Lambda for real-time inference. Transcripts are passed to Amazon Bedrock (Claude/Titan) to generate tailored response suggestions. A custom-built Data Loss Prevention (DLP) and threat detection engine runs in parallel—using both regex and GenAI to catch PII leaks, credential mentions, or signs of social engineering.

Built With

Share this project:

Updates