Coach Commercial AI

Coach Commercial AI is a complete sales coaching ecosystem built to evaluate commercial calls, guide sales representatives, and centralize managerial supervision. The platform combines SalesIQ for the web administration experience, CoachMate for the mobile commercial experience, a NestJS microservices backend, FastAPI AI services, and a RAG agent for document-based assistance.

The strongest part of the project is the AI evaluation engine. A recorded call is uploaded, processed asynchronously by Celery, transcribed with Gemini 2.5 Flash, diarized by speaker, enriched with talk-ratio metrics, then evaluated against a dynamic sales methodology managed from the catalog service. The result contains a global score, detected pipeline steps, detailed sub-criteria justifications, transversal metrics, red flags, strengths, weaknesses, improvement axes, and coaching recommendations.

The system also includes a RAG knowledge base for company documents, a mobile CRM for field sales activity, benchmark workflows to compare AI scoring with human annotations, activity logging, AI usage tracking, observability, and Docker Compose deployment for the complete distributed environment.

• Built a full AI-powered sales coaching ecosystem composed of a web admin platform, mobile app, backend microservices, evaluation engine, and RAG assistant.
• Implemented secure authentication with admin/commercial roles, OTP verification, refresh tokens, password reset, failed-login protection, account locking, and activity logs.
• Developed SalesIQ dashboards for commercials, evaluations, grids, targets, documents, benchmarks, recent activity, and operational indicators.
• Created a dynamic evaluation catalog with templates, version snapshots, ordered sales steps, weighted sub-criteria, target profiles, Bloc B transversal metrics, and Bloc C red-flag configuration.
• Implemented asynchronous audio evaluation with FastAPI, Celery, Redis, Gemini 2.5 Flash STT, speaker diarization, language detection, talk-ratio calculation, progress tracking, and retry/re-evaluation workflows.
• Designed detailed evaluation outputs: global score, detected pipeline steps, step-by-step scores, transversal scores, transcription, strengths, weaknesses, improvement axes, coaching recommendations, and red flags.
• Added human benchmark samples and concordance reporting to compare AI evaluations against manager annotations and improve trust in the scoring process.
• Implemented a RAG document module with uploads, versions, soft deletion, indexing status, retry indexing, HMAC-secured webhooks, parsing, chunking, Gemini embeddings, Qdrant storage, hybrid retrieval, reranking, and grounded answers with sources.
• Built the CoachMate Flutter mobile app with GetX state management, secure token storage, contacts, device-contact sync, meeting calendar, evaluation upload, history, details, assistant chat, multilingual UI, light/dark theme, and profile settings.
• Dockerized the complete environment with health checks, Redis, Qdrant, PostgreSQL-backed services, FastAPI workers, Flower monitoring, resource limits, AI usage logs, Langfuse tracing, and Sentry-ready observability.

SalesIQ

Web administration platform for authentication, dashboards, evaluations, score details, coaching grids, documents, commercial management, and activity monitoring.

CoachMate

Mobile CRM application for sales representatives with contacts, calendar, evaluations, assistant access, and day-to-day field activity tracking.