AI for Sales

(AI4S) Workshop
Japan, 27 March, 2024

PhD. Kieu My - CTO RainScales, US

[email protected]
1. Introduction AI
AI market
Fundamental concepts
AI branches
AI products in life

2. Road map AI project

How to approach AI projects
How to present AI to customers
Requirements for data

3. Research and evaluation AI problem

Advanced concepts in AI
Evaluation metrics in AI
Resources to develop AI
References and open source

4. Read more
 Chief Technology Officer (CTO)
PhD in AI from Florence University, Italy.

PhD. Kieu My - CTO RainScales, US Introduction AI 3/40

 Technical Team

AI Engineers
15 talented AI Engineers from top University in Viet Nam
(HCMC University of Technology,... ) and abroad (Germany,...),
experienced from VinAI, FPT and other companies (TMA
Solutions, BAP adventures,...)

Full-stack
60 talented full-stack Engineers (Front-end, back-end, web, app,
RPA) on Python and Java, more than 10 years experiences.

PhD. Kieu My - CTO RainScales, US Introduction AI 4/40

 AI market

AI market revenue worldwide 2022 - 2032 (Billion dollars).

Source: Bloomberg Intelligence

PhD. Kieu My - CTO RainScales, US Introduction AI 5/40

 What is Artificial Intelligence - AI?

How to teach a child to know a dog or cat?

How to teach a child to recognize (determine, distinguish) a dog

or cat?

PhD. Kieu My - CTO RainScales, US Introduction AI 6/40

 What is Artificial Intelligence - AI?
• Integrate human intelligence into computer –> learn,
remember, predict,...
• Training Neural Network to learn –> solving problems.

PhD. Kieu My - CTO RainScales, US Introduction AI 7/40

 Fundamental concepts in AI
Neural Network - NN
Matrices simulate human’s biological neural system.
3 levels: Artificial NN (ANN) –> Deep NN (DNN) –>
Convolutional NN (CNN).

Machine Learning
Technique, procedure for training NN on big data, –> results to
solve problems.

Deep Learning
NN with more layers (deeper layers) and many machine
learning algorithms to train Deep Neural Network.

PhD. Kieu My - CTO RainScales, US Introduction AI 8/40

 Factors impact AI industry?.
PhD. Kieu My - CTO RainScales, US Introduction AI 9/40
 AI branches

Natural Language Processing - NLP

Voice, Sound, Text. Ex: Voice Recognition, Noise cancellation,
Speech to Text, Text to Speech, Translation, Chat,..

PhD. Kieu My - CTO RainScales, US Introduction AI 10/40

 AI branches
Computer Vision - CV
Images, Videos: Recognition, Object detection, Segmentation,...

PhD. Kieu My - CTO RainScales, US Introduction AI 11/40

 AI products in life

Source: leewayhertz.com)

PhD. Kieu My - CTO RainScales, US Introduction AI 12/40

 Some AI products

Optical Character Recognition - OCR

Extracting text information from images of prescription (Image
–> Text)

DeepFake - CV
Everybody Can Make Deepfakes Now!

Vast products in life

Please see in 4 at the end of this slide.

PhD. Kieu My - CTO RainScales, US Introduction AI 13/40

 How to approach AI project
Collecting requirements:
1 Determine if AI can solve this problem? (requirements,
goal).
2 Evaluate the capacity to handle the project.
3 Recognize the correct AI field (CV, NLP,...)
How to make questions:
1 Data (type of existing data, quantity and quality).
2 Current manual process? (step by step from data to
output).
3 Describe all rules? how many rules?
4 Risk and disadvantages: deployment, from users view,
technical challenges, privacy preserving concern.
5 Resources: human, technical, financial.
PhD. Kieu My - CTO RainScales, US Road map AI 14/40
project
 How to approach AI project

AI training:
1 Analyze how many steps required to solve the problem.
2 Determine how many AI models needed?
3 Label data: Annotate for each kind of data for each AI
model.
4 Training AI model, select top SoTA model, architechture.
5 Evaluation training process.
6 Optimization: fine-tuning parameters.
7 Go back step 5 until to Testing step.
More new data –> go back step 3.

PhD. Kieu My - CTO RainScales, US Road map AI 15/40

project
 Steps after finished training AI
Testing
Inference: Test on independent dataset (test set): accuracy,
speed (s/ms for each data / batch data), on certain hardware.

Deployment
• Integrate on edge devices (Jetson Nano, camera, cloud,...).
• API: apps, website,...

Infrastructure
• Training: At least GPU (3090, 4090, A100, H100,...)
• Testing: Only AI model weight (matrix).
• Cloud services: AWS, GCP, Microsoft, data center,...

PhD. Kieu My - CTO RainScales, US Road map AI 16/40

project
 How to present AI to customers
Approach customer
• Using super simple language.
• Showing benefits.

Recognize AI can apply for the problem

• Evaluate requirements and goals.
• Looking for data and samples.

Determine the value of AI solution

• ROI: Timing, investment. Ex: improve productivity,
increase revenue, cut the cost.
• Increasing customers’ satisfaction.

PhD. Kieu My - CTO RainScales, US Road map AI 17/40

project
 Requirements for data (Big Data)

Important role of data

• Diversity and representation.
• Quality and reliability.
• Quantity.

Data collection and Processing

• Data collection: Reliable source and diversity.
• Processing data (pre-processing and post-processing): Noise
cancellation, missing data and normalization.

PhD. Kieu My - CTO RainScales, US Road map AI 18/40

project
 Advanced concepts in AI
3 learning models of AI
1 Supervised Learning: go with annotated data (labeled).
2 Unsupervised Learning: Clustering.
3 Reinforcement Learning: learning with reward such as
gaming, autonomous driving.

Advanced concepts
• Overfitting: too complicated AI architecture, few data.
• Underfitting: AI architecture too simple, data too difficult.
• Validation and Cross-Validation: train, test, valid (k-fold).
• Fine-tuning: pre-train, optimization.
• Generative AI: Music, images, video, text (ChatGPT),...
PhD. Kieu My - CTO RainScales, US Research and 19/40
evaluation AI
problem
 Common AI problems

Classification, Classification + location, object detection,

semantic segmentation, instance segmentation,...

PhD. Kieu My - CTO RainScales, US Research and 20/40

evaluation AI
problem
 Evaluation metrics in AI
Metrics to evaluate results
• Each problem has different metric.
• Ex: Detection: True Positive (TP), True Negative (TN),
False Positive (FP), False Negative (FN).
• Accuracy: How AI model is correct overall (corrected
prediction on a whole dataset).
• Precision: How AI ML model is correct when predicting the
target class (accurate of this AI model for this data).
• Recall: How AI model can find all objects of the target
class (correction of this AI model compared to labeled
correction).
• mAP (Mean Average Precision): For many classes

PhD. Kieu My - CTO RainScales, US Research and 21/40

evaluation AI
problem
 Evaluation metrics in AI

PhD. Kieu My - CTO RainScales, US Research and 22/40

evaluation AI
problem
 Evaluation metrics in AI

PhD. Kieu My - CTO RainScales, US Research and 23/40

evaluation AI
problem
 Resources to develop AI
Infrastructure
• High performance: Real-time, complex calculation.
• Storage big data, flexibility, super-fast connection.

Hardware
• GPU (Graphics Processing Unit, TPU (Tensor).
• Edge devices (Jetson Nano,...)

Software
• Framework: Pytorch-Facebook, TensorFlow-GG, keras.
• scikit-learn, pandas,...

PhD. Kieu My - CTO RainScales, US Research and 24/40

evaluation AI
problem
 References and open source

Free open access to top-tier conferences

• CVF. CVPR - Computer Vision and Pattern Recognition
• ECCV - European Conference on Computer Vision
• ICCV - International Conference on Computer Vision
• ICML – International Conference on Machine Learning
• NIPS - Neural Information Processing Systems

Open source for SoTA AI models

• Github, Jupyter Notebook, Google Colab.
• Paper with code

PhD. Kieu My - CTO RainScales, US Research and 25/40

evaluation AI
problem
 Q&A

Q&A

PhD. Kieu My - CTO RainScales, US Research and 26/40

evaluation AI
problem
 Read more

AI USE CASES & APPLICATIONS ACROSS MAJOR

INDUSTRIES

PhD. Kieu My - CTO RainScales, US Read more 27/40

 Healthcare
• Medical imaging: X-rays and CT scans, MRI,...
• Personalized treatment: predict disease risks, detect
irregular patterns.

Transportation
• Risk prediction: Monitoring and prediction.
• Safety: warning, manage.

PhD. Kieu My - CTO RainScales, US Read more 28/40

 Retail and e-commerce
• Personalized shopping experience: customer behavior,
preferences, and purchase history, offering tailored product
suggestions.
• Inventory management and demand forecasting: AI
optimizes inventory levels by predicting demand patterns
• Chatbots for customer service: AI provides instant support,
and assists with order tracking.

PhD. Kieu My - CTO RainScales, US Read more 29/40

 Security
Detection, monitoring, prediction in public places, violent.

Military
UVA, drone: supporting dangerous tasks.

PhD. Kieu My - CTO RainScales, US Read more 30/40

 Food tech

• Supply chain optimization: AI optimizes the food supply

chain by predicting demand, managing inventory, and
improving logistics.
• Quality control and inspection: Computer vision algorithms
can detect defects, contamination, or irregularities in food
products.
• Flavor and recipe optimization: Food companies use
machine learning algorithms to create new and innovative
recipes that align with changing consumer tastes.

PhD. Kieu My - CTO RainScales, US Read more 31/40

 Banking and financial services

• Fraud detection and prevention: ML identify anomalies,

and flagging potentially fraudulent activities.
• Credit scoring and risk assessment: Credit scoring powered
by AI utilizes alternative data sources.
• Algorithmic trading and robo-advisors: AI algorithms are
used in algorithmic trading to analyze market trends,
execute trades, and optimize investment portfolios.

PhD. Kieu My - CTO RainScales, US Read more 32/40

 Supply chain and logistics

• Tracking and monitoring shipments: real-time tracking and

monitoring of shipments, offering timely alerts for any
delays or issues, optimizing delivery routes based on
location and traffic patterns.
• Automated inventory management: AI-powered inventory
management enhances warehouse efficiency.
• Optimization of warehouse layout and space utilization:
logistics enhances warehouse operational efficiency by
analyzing data on product demand and inventory levels.

PhD. Kieu My - CTO RainScales, US Read more 33/40

 Travel

• Booking assistance: offering details about pricing, helping

compare options that align with their preferences,
schedules, and budgets.
• Personalized itineraries: AI empowers travel companies to
create personalized itineraries based on individual
preferences. By considering budget, time constraints, and
social media interests.
• Trip planning: recommendations based on the traveler’s
preferences and past travel history. AI can aid in
suggesting popular attractions, restaurants, and activities.

PhD. Kieu My - CTO RainScales, US Read more 34/40

 Real estate
• Automated property valuation: AI algorithms determine
real estate worth by considering historical data, local
market conditions, and property attributes.
• Portfolio optimization in real estate investment: AI
maximizes real estate portfolio returns by utilizing machine
learning to analyze vast datasets, predicting property
values, rental income, and market fluctuations.

Media and entertainment

• Music recommendation systems: AI utilizes user data,
including listening history and preferences, to craft
individual profiles (Spotify).
• AI in game design and playing: creating realistic and
challenging levels that enhance the player’s experience.

PhD. Kieu My - CTO RainScales, US Read more 35/40

 Manufacturing

• Defect detection: AI elevates defect detection in

manufacturing by incorporating self-learning capabilities
into conventional systems, thereby diminishing false alarms
and time-consuming manual checks.
• Real-time monitoring and analysis: AI enables real-time
monitoring of production lines, collecting data from various
sources. By quickly detecting anomalies and deviations,
manufacturers can take immediate corrective actions,
optimize processes, and improve efficiency.

PhD. Kieu My - CTO RainScales, US Read more 36/40

 Automotive
• Intelligent traffic management: analyzing live traffic data
from GPS and surveillance cameras, optimizing traffic flow,
alleviate congestion, and improve road safety. Adaptive
traffic signal control systems dynamically adjust signal
timings based on real-time traffic conditions, mitigating
bottlenecks.
• Autonomous driving: cars, trains.

PhD. Kieu My - CTO RainScales, US Read more 37/40

 Education
• Personalized learning: AI tailors content and learning paths
to individual student needs, fostering engagement and
optimizing learning outcomes.
• Intelligent tutoring: tutoring systems offer personalized
guidance and feedback, elevating the learning experience
for students.

Fashion
• Styling and visual merchandising: AI suggests product
combinations and outfit ideas and even enables virtual
try-ons.
• Design and product development: AI tools integrated into
the design and development process, offering design
suggestions, optimizing patterns for material efficiency, and
creating virtual prototypes.

PhD. Kieu My - CTO RainScales, US Read more 38/40

 Information technology
• Software development and testing: AI automates repetitive
tasks in software development.
• IT support and help desk automation: AI-driven chatbots
and virtual assistants streamline IT support by efficiently
handling user inquiries and technical issues.

Hospitality
• Guest sentiment analysis: analyze customer reviews, social
media mentions, and feedback to extract sentiment and
identify areas for improvement.
• Language translation: AI-powered language translation
services can bridge communication gaps between guests
and staff.

PhD. Kieu My - CTO RainScales, US Read more 39/40

 THANK YOU

PhD. Kieu My - CTO RainScales, US Read more 40/40