devaprasath4/caretta

Purpose and Entry Points

Caretta is a real-time sales-call intelligence system: it captures conversation audio/transcripts, detects technical questions or objections, generates AI coaching and answers, and preserves call history. The Node backend starts through server/src/server.js and configures the Express application in server/src/app.js. Static browser-facing entry pages include client/index.html, client/login.html, and client/pages/dashboard.html.

Browser Client

The frontend is plain HTML/CSS/JavaScript rather than a framework application. Dashboard and call-facing behavior is organized in client/js/dashboard.js, client/js/ai-coach.js, and client/js/transcript.js; session authentication is handled by client/js/login.js. client/js/socket.js is the client boundary for real-time server communication, allowing generated guidance and transcript events to reach an active sales screen.

HTTP, Authentication, and Session Boundary

Express exposes separate route domains for identity, calls, transcripts, AI, and knowledge-base operations: server/src/routes/authRoutes.js, server/src/routes/callRoutes.js, and server/src/routes/aiRoutes.js. Authentication controller logic resides in server/src/controllers/authController.js; protected requests pass through server/src/middleware/authMiddleware.js. JWT configuration is implied by the required JWT_SECRET, separating user/session access from call and AI workflows.

Real-Time Call and Transcript Flow

A call session is created and managed through server/src/controllers/callController.js and persisted as server/src/models/CallSession.js. Speech transcription enters through server/src/services/whisperService.js, with transcript request handling in server/src/controllers/transcriptController.js and durable transcript records in server/src/models/Transcript.js. Socket infrastructure in server/src/config/socket.js distributes live call/transcript and coaching updates to connected clients.

AI Coaching Pipeline

Transcript/question content is passed to the LLM boundary in server/src/services/llmService.js, intended for Google Gemini. server/src/services/coachingService.js turns model output into sales-oriented responses, objection guidance, and follow-up recommendations. Suggestions are represented by server/src/models/Suggestion.js; this provides a persistence boundary between ephemeral real-time advice and historical call insights. Voice output can be delegated to ElevenLabs through server/src/services/ttsService.js.

Retrieval-Augmented Knowledge Layer

The repository supports grounding AI responses in product knowledge. Knowledge content is modeled as chunks in server/src/models/KnowledgeChunk.js, with source data at server/src/data/knowledgeBase.js. Retrieval and response augmentation are implemented in server/src/services/ragService.js, while knowledge/AI-facing endpoints are defined in server/src/routes/kbRoutes.js, server/src/routes/ragService.js, and server/src/routes/aiRoutes.js. This is the boundary between internal sales/product knowledge and external LLM generation.

Data, Infrastructure, and External Services

MongoDB is the system of record, initialized via server/src/config/db.js, for users, calls, transcripts, knowledge chunks, and suggestions; user records are defined in server/src/models/User.js. Required runtime services are Node.js, MongoDB, Google Gemini, speech-to-text integration, ElevenLabs, and Socket.IO-style real-time transport. Environment-dependent credentials and connection settings are documented by server/.env.example; centralized error handling is at server/src/middleware/errorMiddleware.js.