Mohammedkh97 / SentyChat

SentyChat

Multilingual Communication Analytics Platform

A unified analytics system for WhatsApp chats and voice calls in mixed Arabic dialects and English. Uncovering actionable insights through cutting-edge AI.

Multilingual

Arabic Dialects & English

Multi-Channel

WhatsApp & 3CX Voice

AI-Powered

Whisper, GPT-4o, RoBERTa

Real-Time

Automated Data Fetching

System Architecture

Seamlessly processing omni-channel interactions through a centralized routing and specialized AI analysis pipeline.

3CX Platform

Voice Calls (.wav)

Data Storage

MongoDB

Call Logs & Timestamps

Odoo DB

WhatsApp Chat Data

Central Router

Orchestrator & Type Detection

Chat Call Analysis

Extracts sentiment, intents, and bilingual summaries from WhatsApp.

Combined Analysis

Merges insights across channels, scoring similarity & unified reporting.

Voice Call Analysis

Transcribes via Whisper, detects medical specialty inquiries.

AI-Powered Features

Comprehensive analytics leveraging advanced LLMs and fine-tuned models.

Voice Transcription

OpenAI Whisper Large-v3 with GPT-4o refinement for Arabic dialects & code-switching.

Sentiment Analysis

Fine-tuned XLM-RoBERTa 5-class classification (positive, negative, neutral, incomplete, unanswered).

Communication Flow (UCF)

Detects conversation phases (Greeting, Intent, Service, Conflict) and specialty inquiries.

Response Analytics

Per-role statistics tracking agent vs. client wait times and first contact detection.

Bilingual Summarization

Dual-language context-aware summaries keeping vital details like names, dates, and costs.

Exploratory Data Analysis

Label distribution, message length, and deep conversation pattern visualizations.

Agent Performance Analytics

Deep dive into agent-level metrics, evaluating adherence to communication flows, efficiency, and overall interaction quality.

UCF Adherence

Measures completion of the 8-point communication checklist (Greeting, Intent, CTA, etc.)

% Score

Quality & Soft Skills

Aggregated overall quality score based on Politeness, Empathy, Clarity, and Professionalism.

Out of 10

Resolution Efficiency

Tracks average agent delay and first response time in seconds across all sessions.

Seconds

Derived Outcomes

Calculates Customer Satisfaction (CSAT proxy) and Customer Effort Score (1-5).

Outcome

Technical Stack

Django 5.2.5 Backend
Django REST Framework
PostgreSQL Database
OpenAI Whisper Speech-to-Text
GPT-4o LLM Engine
XLM-RoBERTa Sentiment Model
PyTorch ML Framework
Pandas & NumPy Data Analysis