AI call quality analysis system
Project Goal
Development of a comprehensive system for automating the collection, transcription, and AI analysis of calls from corporate PBXs and online channels. The system provides real-time monitoring of service quality, detection of critical errors, and generation of analytical reports for management.
Key Features
• Automatic call collection from Binotel API — hourly synchronization of completed calls (incoming and outgoing) with filtering of internal numbers
• Uploading and transcription of audio via AssemblyAI — speech recognition with speaker separation (manager / client)
• AI assessment of call quality via OpenAI — three assessment schemes depending on the type of call (incoming, warm outgoing, cold outgoing)
• Three interactive dashboards in Metabase — call analytics, KPIs and performance, quality by ratings
• Admin panel for managing prompts — ability to edit assessment criteria without access to n8n
• Web view of call details — interactive interface with dialogue, rating, metadata, and audio
Technical Details
The system consists of 8 n8n workflows and is built on a microservices architecture:
• Collector (hourly) — request to Binotel API, normalization, and recording in call_queue
• Processor (every 10 minutes) — uploading audio, starting transcription in AssemblyAI
• Transcription (webhook) — receiving the completed transcript from AssemblyAI
• AI Agent (on schedule) — quality assessment via GPT with Structured Output Parsing
• Infrastructure: Docker (Hetzner), PostgreSQL, Qdrant, Metabase, Nginx Proxy Manager with automatic SSL
Advantages of the Solution
• Full automation — no manual processing, the pipeline operates 24/7
• Real-time quality control — instant detection of critical errors by managers
• Detailed analytics — dashboards with filtering by managers, dates, types of calls
• Flexibility of criteria — admin can change prompts and assessment schemes without a developer
• Scalability — architecture is ready for an increase in the number of calls and operations
Results
The system operates in automatic mode. In the first month, over 13,500 calls were analyzed (not including historical data which was also calculated). Dashboards show a clear correlation between service quality and sales results. Management received a tool for objective assessment of team performance.
The final cost of call assessment is $0.003 (approximately 15 kopecks).
#n8n #Binotel #AssemblyAI #OpenAI #Metabase #PostgreSQL #Automation #CallAnalytics #AI #Telephony #UkrainianRetail #Portfolio
Development of a comprehensive system for automating the collection, transcription, and AI analysis of calls from corporate PBXs and online channels. The system provides real-time monitoring of service quality, detection of critical errors, and generation of analytical reports for management.
Key Features
• Automatic call collection from Binotel API — hourly synchronization of completed calls (incoming and outgoing) with filtering of internal numbers
• Uploading and transcription of audio via AssemblyAI — speech recognition with speaker separation (manager / client)
• AI assessment of call quality via OpenAI — three assessment schemes depending on the type of call (incoming, warm outgoing, cold outgoing)
• Three interactive dashboards in Metabase — call analytics, KPIs and performance, quality by ratings
• Admin panel for managing prompts — ability to edit assessment criteria without access to n8n
• Web view of call details — interactive interface with dialogue, rating, metadata, and audio
Technical Details
The system consists of 8 n8n workflows and is built on a microservices architecture:
• Collector (hourly) — request to Binotel API, normalization, and recording in call_queue
• Processor (every 10 minutes) — uploading audio, starting transcription in AssemblyAI
• Transcription (webhook) — receiving the completed transcript from AssemblyAI
• AI Agent (on schedule) — quality assessment via GPT with Structured Output Parsing
• Infrastructure: Docker (Hetzner), PostgreSQL, Qdrant, Metabase, Nginx Proxy Manager with automatic SSL
Advantages of the Solution
• Full automation — no manual processing, the pipeline operates 24/7
• Real-time quality control — instant detection of critical errors by managers
• Detailed analytics — dashboards with filtering by managers, dates, types of calls
• Flexibility of criteria — admin can change prompts and assessment schemes without a developer
• Scalability — architecture is ready for an increase in the number of calls and operations
Results
The system operates in automatic mode. In the first month, over 13,500 calls were analyzed (not including historical data which was also calculated). Dashboards show a clear correlation between service quality and sales results. Management received a tool for objective assessment of team performance.
The final cost of call assessment is $0.003 (approximately 15 kopecks).
#n8n #Binotel #AssemblyAI #OpenAI #Metabase #PostgreSQL #Automation #CallAnalytics #AI #Telephony #UkrainianRetail #Portfolio
Dnepr 
Very busy