TMA AI Newsletter Quarter III of 2025

TMA Document Intelligence accelerates document processing with multi-modal AI, self-learning agents, and business-driven configuration. It allows non-technical users to define logic in natural language, reducing manual effort and improving efficiency.

• Semantic document understanding: AI agents deliver high-accuracy data extraction, going beyond basic OCR
• Smart integration: Generates schemas, maps fields, validates & transforms data
• Runbook logic definition: Defines complex business logic and agent workflows by using natural language and visual tools
• Multi-modal extraction: Extracts structured data from any document type or input format
• Adaptive learning (human-in-the-loop): Evolves AI scales confidently based on human feedback
• Workflow orchestration: Manages and controls the entire document process

• Infrastructure: MongoDB, Redis, PostgreSQL, MindsDB, Kafka & Flink
• Application: React, Nextjs, TailwindCSS, Python FastAPI
• AI technology/framework: LLMs/MLLMs, LangGraph, AI Agent/Multi-Agent, Agentic Automation, MCP
• DevOps: Docker, Gitlab CI

• Accelerated efficiency: Automates manual tasks to boost speed and free your team for high-value work
• Maximum accuracy & quality: Extracts data with up to 100% accuracy, including reliable data validation and transformation
• Actionable data: Turns unstructured documents into accurate, ready-to-use data that drives decisions
• Rapid ROI: Achieves fast returns through business-driven configuration (quick deployment) and seamless integration with existing systems
• Visibility: Full control of the document lifecycle on one platform

This is a voice-first assistant that enhances senior independence with empathetic conversations, health reminders, and 24/7 safety monitoring, bringing peace of mind to families and value to healthcare providers.

• Proactive engagement: Daily chats on health, hobbies, and memories, adapting to user needs
• Personalized routines: Conversational reminders for meds, hydration, and exercise with caregiver alerts
• Voice-first interaction: Natural, real-time voice control with user in charge
• Safety monitoring: Detects distress words and abnormal sounds like falls or shouts
• Emotional insight: Reads mood, adjust tone, and share trend reports with caregivers
• Persistent memory: Remembers personal details while supporting privacy controls

Speech & Language AI: ASR (speech-to-text), TTS (natural voice), NLU (intent/context), LLMs (dynamic dialogue)
Emotion & Sound Analysis: Detects mood and emergency sounds
Edge–Cloud & Integration: Scalable, real-time performance with seamless cross-platform interoperability

• Emotional Well-being & Independence: Reduces loneliness and provide personalized daily routine guidance
• Family Connection: Caregiver notifications and shared updates
• Home & Assisted Living: Daily wellness monitoring with scalable support for multiple residents
• Healthcare Support: Enhances telehealth, track adherence, and keeps families informed and supported

AI Kiosk is a virtual assistant for hospitals that streamlines patient intake and appointment scheduling. It enables faster, more accurate registration, reducing staff workload, minimizing wait times, and improving the patient experience.

• Vietnamese voice recognition for accurate medical data
• AI voice assistance & guided workflow
• Custom AI training with patient emotion detection
• Simple, user-friendly interface for healthcare

• User interaction: SpeechRecognizer API, Text-to-Speech
• Language processing & operations: NLP, AI-Driven UX/Output, medical symptom assessment frameworks (PQRSTU & OLD CARTS), and clinical directory integration
• AI & Integration: T-VA, HIS, and VSS portal via RESTful APIs

• 250% Efficiency Boost / Significantly Reduced Patient Wait Times
• 99.5% Reduction in Nurse workload
• 58% Operational Cost Savings
• Higher Patient Satisfaction / Improved Registration Experience

This solution enables hands-free, voice-based data entry. Using Automatic Speech Recognition (ASR) and Natural Language Processing (NLP), it transcribes and structures clinical observations directly into patient records with high accuracy.

• Speech-to-text for structured medical data
• Automatic classification and mapping to electronic health records (ehr)
• Real-time data validation for clinicians
• Vietnamese dialect support for improved accuracy
• Intuitive interface for easy clinical adoption

• Cross-platform availability: Compatible with Android & iOS ecosystems
• Android/IOS integration: SpeechRecognizer
• AI/ML components:
  • Large language models (LLMs)
  • Document understanding: DeepDoc Analysis
• Device compatibility: Optimizes for ruggedized hardware

• Time savings: Up to 50% reduction in clinical data entry time
• Improved data accuracy: Ensures data integrity and minimizes errors
• Enhanced patient outcomes: Supports timely diagnostics and quality care
• Operational scalability: Seamless HIS integration across facilities

An AI-powered language learning platform that understands your level, adapts to your learning style, and helps you practice with real-world conversations and instant feedback.

• Smart level assessment: Knows where you stand with a smart assessment that identifies your level, gaps, and goals
• Personalized learning path: Learns through lessons that adapt to your style, goals, and level - guided by your performance and progress
• AI conversation role-plays: Practices real-world conversations with AI guidance
• Instant speech feedback: Improves clarity and fluency with real-time feedback on pronunciation, pacing, tone, and rhythm
• AI language and assistant tutor: AI personal coach supports 24/7 with tracking progress, offering tips, and guiding next steps

• Cloud & edge computing: Azure
• AI techniques: Azure Speech Avatar, OpenAI (GPT-4o), OpenAI Whisper, LangChain, FastAPI, Python, Milvus
• Database: PostgreSQL
• Back-end: C#, .NET, Umbraco
• Front-end: ReactJS, Angular, PrimeNG

• Clear assessment: Automates level checks, reducing manual onboarding
• Adaptive learning: Tailors content to goals and learning styles
• Real-world practice: Simulates job-specific conversations with instant feedback
• Instant guidance: Real-time support on pronunciation, tone, and fluency
• Continuous progress: Ongoing AI tutor assistance with smart alerts

An AI job advisor that enhances candidate engagement by scanning CVs, suggesting the right job matches, and answering candidate questions 24/7 – helping HR teams deliver smarter hiring support and a better candidate experience.

• 24/7 candidate support: Provides instant AI-powered responses to candidate questions
• Smart CV scanning & role matching: Automatically extracts CVs key details and suggests best-fit roles
• Seamless application: Auto-fills applications from your CV or chat history, reducing drop-off
• Recruiter dashboard: Centralized hub to manage candidates and update chatbot knowledge

• AI techniques: OpenAI (GPT-4.1 mini), OpenAI Whisper, LangGraph
• Database: PostgreSQL, Milvus
• Back-end: FastAPI, Python
• Front-end: ReactJS

• Engage candidates 24/7: Provides instant answers, improving responsiveness and employer brand perception.
• Smart job matching: Automatically scans CVs and suggests suitable jobs for clearer career choices
• Reduce workload: Offload repetitive Q&A and prescreening tasks to AI
• Improve candidate experience: Guides candidates with personalized suggestions and next steps

The Warehouse Management System (WMS) Agent leverages AI to transform warehouse operations. It enables teams to search inventory instantly, track goods and orders in real time, and generate data-driven insights that enhance accuracy, speed, and decision-making.

• Instant inventory search: Retrieves product details by SKU or description within seconds
• Smart location tracking: Identifies precise storage positions and provides detailed item location data
• Real-time order monitoring: Tracks inbound and outbound orders to maintain visibility across the supply chain
• Automated reporting: Delivers instant reports and alerts, including top-performing SKUs, low-stock notifications, and stock movement summaries

• AI techniques: n8n framework, OCR model, Gemini model, Prompt Engineering
• Front-end: React, Chakra, Redux
• Back-end: Python, FastAPI
• Tool & service: Firebase, Jenkins, Swagger
• Infrastructure: Keycloak, Redis, PostgreSQL, RabbitMQ, Docker, Kubernetes
• Architecture: Microservices 

• Enhanced efficiency: Automates routine warehouse tasks, reducing manual effort and speeding up operations
• Smarter decision-making: Leverages real-time insights and predictive analytics to optimize inventory and order management
• Improved user experience: Provides intuitive interaction via chat, voice, or image queries, making the system easier and faster to use