[maum.ai] 3 Foundation model based Embodied AI products_En V1.0

Embodied AI Products
3 Foundation Modules
Our 3 Foundation Modules are modular solutions developed around our core technologies: the Language Model
(MAAL), Voice Conversation Model (SUDA), and Autonomous Driving Model (WoRV). These modules are designed
for seamless integration across various application environments, including on-premise AI, on-device AI, and
end-to-end autonomous driving. They deliver optimal performance and exceptional scalability to meet diverse
business needs.
Based on MAAL
MAAL demonstrates human-like judgment and natural
responses, even in unexpected questions or situations.
Language Foundation Model ‘MAAL’
(Multilingual Adaptive Augmentation Language-model)
Based on SUDA
Just like a natural human conversation, SUDA
continuously processes input while the other party speaks,
delivering responses at precisely the right moment.
Audio Foundation Model ‘SUDA’
(Seamless Uninterrupted Dialogue Assistant)
Based on WoRV
Leveraging visual data, WoRV makes immediate
decisions, reacts quickly to unexpected situations,
and enables end-to-end autonomous driving.
Vision Language Action Model ‘WoRV’
(World model for Robotics and Vehicle control)
Voice Conversation On-Device AI
An innovative solution integrating STT, LLM, and
TTS, ensures zero-latency communication.
On-premise LLM Appliance
Powered by MAAL-Albatross (70B), it enables secure
deployment of Agent LLM and RAG chatbots without
concerns about data leakage.
E2E Autonomous Driving Kit
By interpreting visual information linguistically, it
analyzes situations and performs appropriate
actions. It is applicable across various industries,
including service robots and cleaning robots.

6 Core products
The 6 core products represent a suite of final solutions built upon our innovative technologies and the
3 foundation modules: the Language Model (MAAL), Voice Conversation Model (SUDA), and Autonomous Driving
Model (WoRV). These modules are seamlessly integrated with hardware to deliver optimized solutions across
diverse environments and requirements. Designed as powerful tools for accelerating digital transformation and
enhancing productivity, they are tailored to meet the needs of enterprises.
Expertise
Multichannel
Yielding
Bridging
Decisive
Operative

On-Premise LLM Appliance
For Enterprises Unable to Use ChatGPT
ChatGPT streamlines repetitive tasks and saves time, significantly enhancing workplace productivity. However, many
enterprises are unable to use it internally due to concerns about 'network separation' and 'data leakage', leading to
significant losses as employees are unable to leverage AI tools.
On-Premise LLM: The Emerging Alternative
On-premise LLM is a large language model deployed on a company’s private servers rather than in a cloud
environment. Operating within the internal network, it ensures complete control over external access, providing robust
protection for corporate data.
Performance Comparable to ChatGPT
To evaluate the capabilities of MAUM.AI's LLM, 'MAAL 70B', we tested its performance using LogicKor prompts
and GPT-4 Turbo scoring, benchmarking it against other leading models.
On-Premise LLM Package
MAUM.AI offers a package that combines a secure LLM chatbot with tools like Chat Play and LLM Task UI to
enhance workplace efficiency.
[Overall Test Scores] *based on 17 commonly used enterprise tasks (Score out of 10)
MAAL 70B 9.06
GPT-4o 9.59
Gemini-1.0-pro 8.31
Task UI
Task UI is a pre-built tool to automate
internal tasks such as content
generation, summarization,
proofreading, translation, analysis, and
coding. It can be tailored to meet the
specific needs of clients.
Chat Play
MAUM.AI's LLM search tool
supports building RAG chatbots
through document uploads, allowing
fast and accurate information
retrieval with a single query.
MAAL Chatbot
With the MAAL chatbot, you can
receive answers to your questions
through seamless, interactive
conversations.
B A S E D

STT·LLM·TTS Integrated On-Device AI
MAUM.AI's on-device AI is an innovative solution that integrates STT (Speech-to-Text), LLM (Large Language Model), and
TTS (Text-to-Speech) to achieve zero latency. This allows voice interactions on compact devices such as smartphones
and kiosks, without the need for an internet connection.
On-Device AI
STT
Speech-To-Text
TTS
Text-To-Speech
LLM
Large Language Model
Successful Implementation of LLM Latency Under 1.0 Seconds
MAUM.AI has successfully reduced the total latency to under 1.0 seconds by fully utilizing NPU (Hexagon)
resources and achieving significant improvements in processing speed and performance through efficient porting.
B A S E D
[LLM Latency]
*STT/Controller Latency Measured in ms; Considered Negligible
CPU&GPU Resource Usage 1.9
100% NPU Utilization 0.7

MAUM.AI's On-Device AI
MAUM.AI's On-Device AI maintains high processing capabilities with minimal resources, making it suitable for a
wide range of industries, including mobile devices, IoT, and smart appliances. Notably, it enables device control
without an internet connection, ensuring stability and convenience. It allows users to control home appliances via
voice commands and offers personalized services by learning user behavior. Additionally, the hybrid connection
with the cloud allows simple commands to be processed locally, while complex tasks are handled through cloud
integration, maximizing flexibility and accuracy.
Use Cases of On-Device AI
We create AI products that independently perform various tasks through voice interactions. The LLM understands
customer intent, enabling seamless control of various device functions and facilitating natural conversations. Even in
cases of minor speech recognition errors, the system comprehends the context, ensuring smooth understanding and
functioning as a highly capable agent.
Control Without
Internet Connection
Personalized Service
Through User
Behavior Learning
Enhanced Flexibility
and Accuracy
Hybrid Connectivity
with Cloud Integration
Smart Home & Appliances with
seamless voice control
Kiosks enabling
voice-based communication
AI Service Robot

E2E Autonomous Driving Kit
What is the E2E Autonomous Driving Kit?
The E2E Autonomous Driving Kit is a core hardware solution for AI-based visual information processing and
end-to-end autonomous driving, consisting of the NVIDIA Jetson Orin and four cameras. This kit is equipped with
WoRV (World Model for Robotics and Vehicle Control) technology, enabling it to linguistically understand the
environment and make immediate decisions, ensuring stable driving in various situations.
Ability to Handle Exceptional Situations
WoRV interprets visual information linguistically, allowing it to accurately
recognize and avoid exceptional situations such as glass walls or puddles.
This capability ensures safe driving even in complex environments.
Ability to Understand Commands
Traditional autonomous driving systems often fail when variables interfere,
even with precise route instructions. WoRV, however, can understand
natural language commands, and execute ambiguous instructions like
"Spray pesticide between the trees," offering flexible and intuitive driving.
Adaptability to Environmental Changes
WoRV can navigate with simple maps or LiDAR scan data and can avoid
obstacles that are not present on the map by processing real-time camera
feeds. This adaptability allows the system to quickly adjust to changes in
the environment, supporting efficient driving.
Equipped with Jetson Orin, the
Ultra-Compact High-Performance
Computing Platform
360-degree Environment
Detection / Complete
Visibility with Four Cameras
Integrated with WoRV
B A S E D
AI Call Center
How WoRV Differs from Existing Solutions

Use Cases and Future Potential of the E2E Autonomous Driving Kit
The E2E Autonomous Driving Kit offers innovative potential across various industries. In agriculture, it can be used as
an autonomous farming vehicle that navigates between trees and sprays pesticides. In logistics, it can operate as an
autonomous robot that performs smoothly in complex environments, providing both efficiency and safety.
Additionally, in the military, it can be deployed as a reconnaissance robot or landmine detection drone, safely
conducting operations in hazardous areas. Looking ahead, the E2E Autonomous Driving Kit is expected to play a
crucial role in smart cities and mobility services.
지시 이해 능력
기존 자율주행 시스템은 정확한 경로 지시에도 변수로 인해 주행에 실패하는 경우가
많습니다. WoRV는 자연어 명령을 이해해 "나무 사이로 농약을 뿌리고 와"와 같은
모호한 지시도 수행하며, 유연하고 직관적인 주행을 제공합니다.
Modular and versatile scalability
WoRV modularizes autonomous driving technology, allowing it to operate
freely across a variety of hardware platforms. This enables highly scalable
and compatible applications, regardless of the environment.
Location Tracking Accuracy
WoRV utilizes Odometry technology to continuously correct its position in
real time. By leveraging camera inputs, it ensures accurate location tracking,
solving drift accumulation issues and supporting stable driving.

Expertise
Expert LLM
The language foundation model 'MAAL'-based on-premise
chatbot package is a professional office AI solution that
operates on both PC and mobile. It performs a wide range
of expert roles and integrates seamlessly with backend
systems, enhancing productivity in diverse fields such as
consulting, beauty, and education.
Labor Inspector, Legal Advisor, Beauty Assistant, Banker, IR
Officer, Hospital Consultant, English Tutor, Korean
Instructor, Game Guide, Simultaneous Interpreter
Expertise AI Solutions
Multichannel
AI Call Center
Voicebots and chatbots seamlessly integrated with
telecommunication networks automate tasks like outbound
telemarketing, follow-up calls, and inbound reservations.
These AI solutions enhance customer service efficiency by
providing inquiries and reservation processes across
platforms such as websites and social media applications.
Outbound Telemarketer, Happy Call Consultant, Inbound
Reservation Officer, TMQA
Multichannel AI Solutions
Bridging
Voice Conversation Display-Based Guide AI
Kiosks equipped with voice recognition and AI-powered
human interaction technology automate front-desk tasks,
including hotel check-ins, airport navigation, department
store guides, and university administrative services. Integrat-
ed with payment and scanning capabilities, these kiosks
deliver fast, efficient, and seamless customer experiences.
Hotel check-in tasks, airport facility guidance, department
store floor guidance, university academic administration
guidance, announcers, weathercasters
Briding AI Solutions
6 Core Products Types

Operative
Robots and Vehicles with the
Autonomous Driving Model WoRV
Special-purpose autonomous vehicles and AI robots enhance
efficiency with 24/7 operations in industries like agriculture,
cleaning, and logistics. Autonomous driving AI technology is
actively applied to tasks such as goods transportation, pesticide
spraying, tangerine harvesting, building cleaning, and last-mile
delivery.
Pesticide-spraying agricultural machinery, building cleaning
robots, last-mile delivery robots
Operative AI Solutions
Decisive
Robots and Vehicles with
Voice Conversation and Autonomous Driving
Robots equipped with autonomous driving units and expert
LLM serve as gallery docents and sales assistants in offline
stores. By integrating autonomous mobility and voice
conversation AI, they efficiently automate guidance and sales
tasks.
Gallery docents, offline store salesperson
Decisive AI Solutions
Yielding
AI Robots with Voice Conversation,
Autonomous Driving, and Robotic Arms
Robots combined with autonomous mobility, robotic arm
manipulation, voice conversation, and expert LLM automate
various tasks performed by humans, including caregiving in
nursing homes and serving in restaurants. These robots
deliver high precision and efficiency, operating 24/7 to move,
converse, and execute intricate tasks seamlessly.
Citrus picking machinery, nursing home caregiver,
serving robot
Yielding AI Solution
6 Core Products Types

Device Implementation
Web Implementation
AI-Powered Labor Inspector
Support System
Your AI Partner for Labor Law Solutions for Both Workers and Employers
E type
Supporting Better Work Environments
The AI Labor Inspector Support System leverages RAG(Retrieval-Augmented Generation)-based LLM(Large
Language Model) technology to provide optimized solutions for both workers and labor inspectors. For workers,
this system offers accurate labor law information quickly and easily via 24/7 chatbot support, helping resolve
complex issues efficiently. For Labor Inspectors, it streamlines tasks such as document summarization, analysis,
search, and proofreading through intuitive, conversational interactions, enhancing overall work productivity.
Online Filing Petition Assistance
Simplifies petition form submissions by guiding
users step-by-step through an interactive Q&A
process.
AI Labor Law Knowledge Search
Provides solutions to labor law queries that are
often difficult to ask, offering tailored guidance
on various issues such as severance pay and
dismissals.
Barrier-Free Kiosk Submission
The voice-recognition kiosk allows both disabled
and non-disabled individuals to submit complaints
seamlessly. Even those unfamiliar with online
systems can resolve issues with ease.
Conversation with Documents
Assists labor inspectors in summarizing, analyzing,
searching, and annotating large volumes of
documents with ease. All documents are
processed securely within the internal network,
eliminating any security concerns.
Labor law provisions can be searched
effortlessly, enabling users to find key pages
and critical content from manuals or docu-
ments in seconds.
Reference Material Search

AICC
Cutting-Edge AI Call Center Consultation System
M type
Use Cases of AICC
Voice Bot and Visible Voice Bot Solutions
The Voice Bot and Visible Voice Bot solution is a product developed by MAUM.AI, integrating our STT, TTS, and
conversation management AI engine, along with a call switch and web server.
Premium Calculation Consent Call
The VoiceBot performs premium calculation consent calls for outbound sales database acquisition in automobile
insurance. It handles residual data tasks that could not be completed by agents, expanding data utilization
efficiency and facilitating the insurance development authority's compliance checks for customer
communications. This system has shown a success rate of 20-30%.
Insurance Company VOC Complaint Analysis
By using LLM (Large Language Model) and TA (Text Analysis) technologies, this solution summarizes VOC
complaints, analyzes complaints and demands, extracts keywords, and classifies complaint types, significantly
enhancing the efficiency of complaint analysis tasks.
• The Voice Bot directly manages customer consultation. Real-time STT, TTS (speech synthesis), conversation
scenarios, and intention classification solutions are integrated with the call system to process interactions.
• Initially, this solution was applied to simple counseling tasks in the financial sector. The system is gradually
being used for various tasks, including customer support in call centers across the retail, manufacturing,
and service industries.
AICC, AI Call center
AICC (AI Call Center) is a next-generation solution that leverages AI technology to maximize customer support
efficiency while achieving cost savings and enhanced customer satisfaction. The AI-based consultation system
covers a wide range of tasks, from resolving customer inquiries and managing issues to supporting marketing and
sales activities, all of which contribute to a revolutionary improvement in customer experience.
AICC 인공지능 콜센터
AI 기술이 인간의 노동을 가장빠르게대체할직업 중의하나가
콜센터상담원이라고 예상되었고,국내에서는2016년도부터 금융권,통신사의
콜센터에 음성인식, 텍스트 분석등의 기술을 활용해서상담원품질 관리나
고객의민원,이슈를분석해왔습니다.
2020년코로나사태로인한 비대면 상담요구가증대하면서,
고객 접점영역에서 AI기술을활용한고객 응대가 확산되고있습니다.
한편,상담원신규 인력공급이어려워지면서,
AI 콜센터의구축은이제 선택이 아닌필수로변화되고 있습니다.
음성봇, 보이는 음성봇 솔루션
음성봇, 보이는 음성봇 솔루션은마음AI 가 자체개발한STT, TTS,대화 관리AI 엔진과 콜 연동Switch,웹서버가통합된제품입니다.
마음 AI는 AICC의 대표적인제품인음성봇을 2019년도부터자체 기술로 개발하였습니다.전화기에 해당하는Maum-Switch, 상담원에해당하는
음성인식,음성합성,대화 시나리오AI 엔진이,실시간으로고객 정보를 확인하고상담할수있도록한시스템입니다.최근에는 고객이 스마트폰의 화면을
보며 대화하는‘보이는 음성봇’을 클라우드서비스로 제공하고있습니다.
고객과의 상담업무를 Voice Bot 이 직접수행합니다. 실시간 STT,TTS(음성합성), 대화시나리오,의도 분류등의 솔루션이콜시스템과 연계하여처리합니다.
초기에는 금융사 단순안내 상담업무에우선적으로적용되었으나,점차 다양한 업무처리 및 유통/제조/서비스 산업 의콜센터상담에활용되고 있습니다.
Customer Omni-channel
Image and
Document Recognition
Agent Call Routing
Agent Call Routing
Cognitive and Language
Intelligence-Based Document Classifier
LLM
Advanced
RAG
AI Agent
Domain
Optimization
Training
FAQ
Support Chatbot
Voice Bot
Chat Bot
Scenario and Learning
Management Monitoring
Statistics Admin
Providing a Stable
AI Service
Operating Environment

AI Docent : LARA
MAUM.AI offers 2 types of AI Docent services: LARA & AIden
Docent LARA (Localized Artificial and Robotic Assistant)
featuring an input-output separation method
output
input
LARA uses voice or touch inputs to deliver results generated by the LLM or AI Human through an
output section.
Ideal for scenarios where the docent conveys a large amount of information, particularly in
noisy environments.
‘LARA in 2024 Experience Center of the 2024 Gangwon Winter Youth Olympics’
Output section can be adjusted in size.
Voice or touch controls for easy use
Establishes unmanned spaces by implementing AI Human M4, powered by 5 core engines.
(Face, Voice, Ears, Eyes, Language)
B type

maum-TOUCH
Tactile Operable Universal Convenient Hearing and Visually accessible
Maum.AI provides barrier-free kiosks to ensure accessibility
With increasing societal demands for accessibility and barrier-free technology, maumTouch is becoming essential.
This Kiosk features 13 functions designed for easy access by individuals with visual, hearing, and physical impairments.
Equipped with Maum.AI’s AI Human M3, maumTouch enables 24/7, year-round automation.
MAUM.AI provides barrier-free kiosks to ensure accessibility
With increasing societal demands for accessibility and barrier-free technology, maumTouch is becoming
essential. This kiosk features 13 functions designed for easy access by individuals with visual, hearing, and
physical impairments. Equipped with MAUM.AI’s AI Human M4, maumTouch enables 24/7, year-round
automation.
Features For Wheelchair Users
•Sensors detect height to adjust the display to eye level
•Patented automatic height adjustment technology
(No. 10-1084-29)
Features for the Visually Impaired (Blind)
•Digital tactile features offered and all visual information
provided in audio
Features for the Elderly or Visually Impaired
(Low Vision/Color Blindness)
•2-level font size adjustment
•High contrast mode
Features for Hearing Impaired Individuals
•All auditory information provided visually
•Alternatives to voice input offered (call button)
Basic Features
Volume Control
(4 levels)
Language Selection
(Korean, English)
Microphone
(Voice Commands)
Manual height
Adjustment Button
Headphone Connector &
Volume Control Key.
B type

AI Simultaneous
Interpreter LINK
Language Interpretation Network Knowledge
Well suited for international conferences, airport screenings, and large shopping centers.
• Multilingual support for major international languages
• Uses specialized terminology contextually
• Operable on-device, allowing offline use
• Compatible with various display devices
LINK delivers seamless, real-time interpretation service with human-like naturalness.
영어 일본어
AGI로 가는 가장 빠른 길, 마음AI의 Foundation_Model
The Fastest Way to AGI, Foundation Model of MaumAI
업종별 대표 기업에게 파운데이션 모델을 만들어드립니다.
We create a foundation model for representative companies by industry.
다국어에 강한 기업용 거대 언어 모델로 번역, 요약, 표, 작성 등에 뛰어난 성능을 보유하며 고객사의 업무 효율성을
획기적으로 제고합니다.
By utilizing a powerful multilingual language model tailored for enterprises, adept at tasks such as
translation, summarization, and generating tables.
다양한 파라미터 사이즈 모델을 보유하여 고객사 각각의 니즈에 따라 맞춤형으로 활용할 수 있습니다.
We possess various parameter sizes models, allowing tailored utilization according to each client’s
specific needs.
MAAL
Automatically corrects typos and redundant phrases
MaumAI provides various artificial intelligence services
based on three foundation models.
USER
마으뭬이아이는 세가지 파운데이션 모델을 기반으로
다양한 인공지능 서비스를 제공하고, 공급합니다
Typo
Repetition
LLM-powered accurate simultaneous interpretation that actomatically corrects typos and
redundant phrases.
B type

Accurate and Safe Pesticide Spraying Machinery
Pesticide Spraying
Agricultural Machinery
Efficient and Unmanned Pesticide Spraying Machinery
Equipped with MAUM.AI's WoRV model, cameras, and sensors, this agricultural machinery ensures precise and
safe pesticide spraying. Its fully autonomous and unmanned operation significantly reduces labor costs while
maximizing work efficiency.
Safe Working
Environment
Unlike traditional speed sprayers, the WoRV-based machinery eliminates the risk of
exposure to harmful pesticides. It can safely and efficiently operate in hazardous
areas such as steep fields and terraced farms, where accidents frequently occur,
providing a safer alternative for workers.
Cost Reduction and
Increased Efficiency
Unmanned operation significantly reduces labor costs. Farmers can schedule
pesticide spraying at their convenience, and the task will be completed
automatically. This eliminates the need for labor-intensive and inconvenient tasks,
such as early-morning spraying, greatly improving the efficiency and convenience
of agricultural work.
Precise Pesticide
Spraying
With WoRV, the machinery recognizes its surroundings in real time and sprays
pesticides with pinpoint accuracy, ensuring efficient and uniform pesticide
distribution.
Eco-Friendly and
Cost-Effective
Electric Machinery
Unlike traditional speed sprayers that relies on kerosene or fuel, the WoRV-based
electric machinery operates on electricity, reducing carbon emissions and
promoting sustainable agriculture. This not only enhances agricultural operations
but also contributes to environmental conservation.
O type
Pesticide Spraying Agricultural Machinery in collaboration with

AI Docent
MAUM.AI offers 2 types of AI Docent services: LARA & AIden
Autonomous Docent AIden
(Artificial Intelligence Docent using Engagement and Navigation functions)
AIden is a 2-in-1 robot that introduces and guides artwork during work hours while also serving as a security
guard after hours.
Autonomous Driving Technology enables AIden to
accompany and converse with individuals, offering
interactive experiences.
Equipped with a LiDAR sensor for mapping, AIden features
advanced spatial detection capabilities.
3D Lidar Sensor
Interact Face Display
Camera & AI Sensor
Interact Front Display
2D Lidar Sensor
Speaker & Mic
UVA & UVC
Ultrasonic Waves Sensor
D type

Language Foundation Model
MAAL
By first understanding the
language , MAAL offers
professional and natural guidance
Audio Foundation Model
SUDA
Seamless & Natural
Conversations
Gilbert
In pure Korean, "Gilbert" means "companion on the road."
Gilbert, powered by 3 foundation models, is a robot designed to observe, think, and act.
Vision Language Action Model
WoRV
End-to-End autonomous driving that
responds appropriately even in
unexpected situations
Drone Cleaning Robot Delivery Robot Security Robot Agricultural Machinery
Gilbert not only processes natural language but also understands contexts, engaging in human-like conversations.
With its language comprehension, Gilbert executes complex, unstructured commands and handle exceptional situations naturally.
MAUM.AI customizes robot Gilbert according to client needs.
With limitless versatility of Gilbert,
Accelerate your company's workflow innovation!
D type

maumBarista : Han-b
maumBarista: The Complete Café Automation System
Beverage Ordering
Convenient beverage ordering is made possible through simple verbal conversations.
The high-performance 'MAAL' engine combined with the Zero-Latency 'SUDA'
ensures accurate user intent comprehension even with slangs or typos.
Beverage Preparation
A robot receives order details and
prepares beverages.
Beverage Serving
The prepared beverage is safely delivered to the customer by the Polaris3D
serving robot, 'e-Re:on.' Equipped with 'WoRV' autonomous driving technology,
e-Re:on performs end-to-end navigation using visual information.
Interactive AI Considering
User Experience
Order process
with a questioning structure
Scenario Construction
tailored to each cafe
Y type

MAUM.AI's Embodied AI Service is adaptable to any profession that
requires voice conversation and expert guidance, including docents,
delivery drivers, security guards, and sales assistants.
MAUM.AI CS : +82-1661-3222 / hello@maum.ai
30, Changup-ro 40beon-gil, 4th Floor, Sujeong-gu, Seongnam-si,
Gyeonggi-do, South Korea (Shiheung-dong, Pangyo IT Center)
https://maum.ai

[maum.ai] 3 Foundation model based Embodied AI products_En V1.0

More Related Content

Similar to [maum.ai] 3 Foundation model based Embodied AI products_En V1.0 (20)

More from Taejoon Yoo (20)

Recently uploaded (20)

[maum.ai] 3 Foundation model based Embodied AI products_En V1.0