TMA AI Newsletter Q1/2024

An application running on the Agent’s PC of the Call Center to capture real-time audio and provide tailored suggestions for customers’ queries, providing suggestions based on a knowledge base to support call agents in answering customer inquiries and concerns.

• Real-time speech recognition: Employ real-time audio streaming during voice calls.  Analyze customers’ voices to understand the questions.
• Integration with company information hub: Integrate with the company knowledge base to provide the relative answer to the customer’s questions.
• Instant suggestions: The system displays corresponding answers relative to customers’ queries.

• Speech-to-text, NER, sentiment analysis
• Natural Language Processing, Deep Learning

• Boosted agent productivity
• Enhanced customer engagement
• Increased sales opportunities

Provide real-time response to user's queries about the applications and common issues when using the system.

• Integration with application information hub: Integrate with the application knowledge base to provide the relative answer to the customer's questions.
• Instant answers: Provide text/ voice answers/ information to customers' questions.
• Multiple language support: Supports English, Vietnamese, etc.

• Generative AI - OpenAI
• RAG document
• Chatbot

• Enhanced user experience
• Optimized operation cost
• Improved customer retention

Analyze and monitor sleep quality based on heart rate data.

• Predict sleep quality based on the curve of heart rate points during sleep.
• Detect if the user's oxygen saturation level drops while sleeping.

• Data processing, data profiling
• Statistics
• Scikit-learn library, classification
• Data visualization

• Assist users in understanding their sleep to adjust their habits before sleep and in daily life.
• Early detect some potential causes and abnormalities during sleep.

AI Solution applies smart cameras to provide real-time surveillance, including weapon detection, trespass alerts, and crowd monitoring, ensuring safety in public areas.

• Real-time weapon detection
• Trespass warning
• Unusual loitering warning
• Breaking for entering indicator and door or window tampering
• Nighttime activity in unusual areas
• Crowd formation and dynamics 
• Aggressive behavior warning
• Vandalism - throwing

• Edge AI
• Object detection, tracking & monitoring
• Behavior analytics

• Real-time detection of potential crimes & violations
• Enhance public safety

Identify, classify and notify road damage to support road maintenance.

• Real-time road damage detection: Utilize Computer Vision technology to identify and classify various types of road damage instantly.
• Map location marking: Pin the detected road damage locations on a map.
• Drivable area segmentation: Divide the road into drivable and non-drivable areas.

• Python, Tensorflow Lite
• Computer vision, Object detection and segmentation

• Real-time detection and reporting
• Efficient resource allocation
• High accuracy

Solution to detect, identify pests and diseases in the field.

• Pest detection: Automatically identify and classify insects in the field.
• Detect diseases in rice paddy: Use images taken from rice plants to determine the type of disease the rice plant is experiencing.
• Real-time notification: Send real-time warnings if the monitoring device detects harmful insects or pathogens on rice plants.

• Computer vision, Object detection
• Object counting/tracking
• IoT devices & sensors, Edge AI
• MLops with Cloud services or On-premise server 

• Flexible deployment
• Human effort optimization in farming work
• High accuracy

Analyze sound to automate durian quality check.

• Sound detection: Develop a specialized tool to tap durian husks and capture the resulting sound feedback.
• Knowledge base: Define the correlations between sound detection and durian ripeness/quality, determining their suitability for export or domestic sale.
• Information portal: Display insights on each durian's ripeness/quality and their readiness for export or domestic sales.

• Spectrogram analysis
• MFCCs, GMMs & HMMs for audio tagging
• PANNs model for audio classification

• Enhanced accuracy
• Optimized operation
• Labor reduction

Apply a sound-based fault detection system to enhance quality control in valve manufacturing.

• Sound detection: Sound sensors capture signals during valve tightening
• Model classification: Data processing and input into a model for fault and normal valve classification
• Notification: Signals are sent to a robot arm to remove faulty products, ensuring only quality items proceed in production

• AI machine Adlink DLAP211, PLC Siemens S7
• Sound analysis
• Model classification

• Enhanced quality control
• Increased efficiency