TMA AI Newsletter Quarter I of 2025

• Multi-agent collaboration: Different agents work together to process tasks
• Adaptive intelligence: AI agents can reason, research, and execute workflows
• Memory & context awareness: Stores historical interactions for better decision-making
• Seamless API integration: Connects with external systems, databases, and enterprise applications

• AI frameworks/technologies: Agentic AI, LangGraph, LangChain, Cloud LLMs, Local LLMs, Computer Vision
• Data frameworks: Kafka, Apache Spark
• Databases: MongoDB, Weaviate, Minio, Redis
• Backend: Python, Spring Boot
• Frontend: NextJS

• Democratized AI orchestration: Empower business users to create AI workflows without coding, reducing technical dependencies and accelerating innovation
• Intelligent process automation: Streamline workflows through collaborative AI agents that reason, research, and execute tasks together
• Enterprise system integration: Connect with existing business systems through robust APIs, enabling AI agents to access and act upon enterprise data
• Adaptable industry solutions: Deploy pre-configured, industry-specific workflows while maintaining the flexibility to customize for organizational needs

Local LLM is an on-premise AI development framework that helps clients quickly build and deploy enterprise GenAI applications. In this new version, this framework support latest LLMs like Meta AI, Alibaba Cloud, Mistral AI, DeepSeek AI

• Fast deployment and scaling: Automate setup, scaling, and high availability
• Flexible model choice: Leverage a diverse ecosystem of open-source AI models for control and customization
• Unified API hub: Access to NLP, Computer Vision, Speech, and LLM models via ready-to-use APIs

• AI models: Various LLMs including LLama 3, Qwen 2.5, Mistral Large 2, DeepSeek-R1, Falcon 3
• Framework: LangChain, Pytorch, LLM, Hugging Face
• Language: Python
• Database: PostgreSQL, MinIO, Qdrant

• Seamless integration: Easily connect with existing systems
• Continuous innovation: Stay updated with the latest AI features
• Rapid development: Build AI apps quickly with user-friendly APIs and pre-built modules
• Secure & reliable: Ensure data security and application reliability
• Cost reduction: Minimize development expenses with automation
• On-premise deployment: Maintain control over sensitive data and ensure compliance with the flexible on-premise deployment options

• Load calculation: Auto-calculate package quantities
• Custom rules: Set specific loading rules
• Optimized loading: Maximize container space with smart algorithms
• Multi-container: Support multiple containers per order
• Clear reports: Generate detailed, easy-to-follow loading reports

• Cloud & Edge Computing: AWS
• AI techniques: DBSCAN, Heuristic Optimization, NumPy, Pandas
• Database: PostgreSQL
• Backend: FastAPI
• Frontend: ReactJS, Chakra UI, Three.js

• Increased efficiency: Reduce manual work with automated processes
• Space optimization: Maximize container utilization with smart algorithms
• Scalability: Handle multiple containers and large orders quickly
• Time savings: Speed up the loading process with optimized planning

• Real-time anomaly detection: Continuously monitor the network to detect anomalies and potential security threats as they occur
• Seamless integration: Easily integrate with existing network monitoring tools
• Advanced analytics & insights: Generate insights and analytics to help administrators understand network behavior and potential risks
• User-friendly dashboard: Provide an intuitive interface for real-time monitoring and security overview of the network

• AI frameworks/technologies: Machine Learning, Scikit-Learn
• Data frameworks: Kafka, Apache Spark
• Backend: Python, Java
• Web UI: Angular

• Faster threat detection & response: Identify anomalies in real-time, allowing network administrators to address issues before they escalate
• Enhanced network security: Provide security overview, helping to prevent cyber threats and unauthorized access
• Improved operational efficiency: Automate monitoring and analysis, reducing the manual workload for network administrators
• Better decision-making: Deliver actionable insights and analytics, enabling informed security and network management decisions

• Unified testing for Web, Mobile, API, CLI, and Desktop applications
• AI-powered OCR and picture detection
• AI self-healing for UI tests
• Automatic test script generation
• Dashboard analytics
• Predictive failure analysis

• AI model: Gemini Flash 2.0
• Language: Python, JavaScript
• Robot-based framework
• Tool/libs: Python libraries, Healenium

• Unified automation framework: All-in-one solution
• Reduce scripting effort: Streamline automation development
• Ensure code consistency: Minimize review time
• Accelerate debugging: Resolve issues faster

• Drowsiness detection: Face & pupil tracking
• Distraction monitoring: Phone use, smoking, drinking, yawning
• SOS signal detection: Hand gestures & threat recognition
• Alert system: Speaker & emergency server

• AI models: Face detection, eyes detection, hand landmark, object detection
• Languages: Python, C++
• Framework: MediaPipe

• Modular design: Easy install & integration
• Competitive pricing: Cost-effective solution
• AI-Powered camera: Enhanced AI safety features
• Direct warning system: Immediate in-vehicle & server alerts

• Real-time fraud detection: Deepfake Detection with fast response (<5s)
• Voice biometrics: Verify caller identity, stop impersonation
• VoIP integration: Real-time, large-scale monitoring

• AI models: Codecfake model, SpeedBrain Classification
• Languages: Python, C++
• Tool/libs: Asterisk, SpeechBrain

• Precision: Detection accuracy exceeding 95%, with a false positive rate below 5%
• Performance: Real-time response with a latency of under 5s with instant alerts to potential scams
• Compatibility: Easily integrate into any VoIP/Call center infrastructure
• Reliable service: Build trust with secure and authentic interactions compared to legacy fraud detection systems

• Sign to text: Part of real-time sign language recognition that converts gestures into written text
• Speech to text: Convert spoken words into text in real time
• Live video and chat interface: Provide live video communication and a chat interface
• Cross-platform access: Available on the web and mobile
• Text-to-sign animation: Translate written text into animated sign language
• Multilingual support: Support various languages to serve a broader range of users

• Deep learning: CNN, Transformer
• Computer vision: MediaPipe, OpenCV
• Backend: Python, Django
• Frontend: ReactJS
• Mobile: Android Kotlin

• Enhanced communication: Bridge the gap between deaf individuals and the hearing community
• Accessibility from anywhere: Allow users to communicate through multiple platforms, anytime and anywhere
• Emergency support: Provide quick and clear communication in urgent situations

• Multiple camera integration
• Video streaming
• Real-time 2D map tracking

• NVIDIA frameworks: TAO Toolkits, Deepstream, Metropolis
• AI frameworks: Pytorch, Tensorflow, Yolo

• Real-time tracking & monitoring
• client behavior analysis
• Enhance security
• Reduce manual labor
• Reduce cost
• Enable various AI applications

• Conversational AI: Engage users with natural, human-like interactions using LLMs
• Lip sync: Match speech with avatar facial expressions
• Speech & emotion recognition: Understand user sentiment for personalized responses
• Multimodal interaction: Support both voice and text input

• AI techniques: Pytorch, OpenCV, Audio, 3D Reconstruction, Speech-to-Text, Text-to-Speech, AI Lip Sync, LLM
• Media server: ossrs
• Database: PostgreSQL
• Backend: FastAPI
• Frontend: ReactJS

• Enhanced user experience: Always ready to assist clients, patients, or employees and respond like humans
• Cost reduction & efficiency: Reduce the need for human client support & training personnel
• Increased engagement & retention: Human-like avatars make interactions more engaging

• Quick information retrieval from enterprise data (PDF, text, Excel, CSV)​
• Automate repetitive support tasks such as FAQs and policy lookups
• Integration with CRM tools like Salesforce​
• Track and analyze user questions to provide service quality insights
• Categorization of chat data by product/portfolio

• Frontend: ReactJS, NextJS​
• Backend: Python, Java SpringBoot, PostgreSQL, MongoDB, Kafka, Redis, K8s​
• AI Backend: LLM Orchestration, RAG-based application, Azure OpenAI LLM

• Increased efficiency: Support users with instant answers from AI assistant
• Easy integration: Connect with existing systems and workflows
• Cost reduction: Lower operational costs through reduced contract rates
• Time savings: Reduce client waiting time and improve service qualit

• Visual agent design: Create assistant flows using a drag-and-drop interface and pre-built components
• Flexible model support: Integrate with leading LLM providers and custom endpoints
• Embedded into any web page: Embed the virtual assistant on any target web pages with APIs and iframe support to perform real-time actions
• Knowledge integration: Synchronize with the knowledge datasource and provide intelligent interfaces for intuitive data management

• AI Models: Support most mainstream model service providers
• Languages: TypeScript, Python, JavaScript​
• Database: MySQL

• Simplified setup and usability​
• Rapid deployment​
• Improved productivity​
• Ease of integration