Technologies We Use in ML and AI Projects

Data Source

OpenML
Kaggle
AWS OpenData
Google DataSet Search
UC Irvine ML Repository
Gemma
Hugging Face
Streaming Datas

Platform

Dedicated Server
Cluster Architecture
AWS
Google Cloud
Microsoft Azure
Google Colab

Framework

Tensorflow
Keras
Scikit
OpenCV
PyTorch
Apache Spark

Software Language

Python
GoLang
Scala

Database

Cassandra
CouchDB
QuestDB
MongoDB
MariaDB

Sectors

As a full-service technology and creative solutions provider, we serve many sectors and industries and work with businesses that are leaders in their fields.

GOVERNMENT
FINANCE
RETAIL
GLOBAL TRADE
MANUFACTURING
ENERGY
TRANSPORTATION

TELCO
AUTOMOBILE
TOURISM
SPORTS
HEALTHCARE
REAL ESTATE

Artificial Intelligence is one of the areas that 4A Labs has been working on for many years.

To take the services we offer to our e-commerce and retail clients to the next level, we have integrated AI-based decision support and measurement tools. Thanks to this integration, our clients can make faster and more accurate decisions, improving their operational efficiency. At the same time, they gain deeper insights into customer behavior through real-time data analysis, allowing them to develop more personalized marketing strategies. These solutions not only boost sales but also significantly enhance customer satisfaction. As a result, our clients increase their competitive edge while achieving more sustainable business processes.

Some of Our Artificial Intelligence Projects

We have added a digital food, health and food assistant to CarrefourSA’s mobile application. The application, which includes features such as recipes, food recognition from images, listing the ingredients of the meal, sorting the necessary ingredients and adding them to the basket, calorie calculation, meal planning, and allergy tracking, is added to the application as a digital assistant via an LLM-based AI model. We will see it in the CarrefourSA application in the markets very soon.

To develop this platform, we scanned approximately 1.2 million food images and contents and ran approximately 1.8 million food and ingredient data on ML models and algorithms.

CuraMuse is your personal shopping expert, combining artificial intelligence and human touch. It analyzes your interests, shopping habits and more to provide personalized recommendations based on your or someone special’s unique preferences. As your virtual shopping assistant, CuraMuse carefully considers every detail and prepares recommendations that fit your lifestyle.

A process management and quality control application based on scikit and pytorch that can be used by all units has been developed. The system, which can take database and excel files as source input, provides insight and makes production estimates. It was developed with machine learning and deep learning models.

Carvak, a global brand in second-hand vehicle purchasing and sales, performs 360-degree vehicle scanning, paint and scratch control over photographs, and vehicle damage detection using machine learning-based image processing algorithms. With this infrastructure, 360-degree inspections and controls of over 5,000 vehicles are performed automatically on the platform every month.

The cloud environments of the websites of Korean makeup brands in Türkiye are controlled by an ML and LLM-based infrastructure as equivalent in Superonline data centers. All transactions are processed on the ML model with logs and metrics received from Prometheus, Datadog and NewRelic, and the metrics are evaluated and transmitted to the relevant personnel as summaries that the IT personnel can understand.

Little Lunches, a healthy food and meal planning app for kids and families, is the most widely used digital health and food assistant in the United States. ML models prepared according to 780 thousand different American and South American food cultures and content are presented to users as an AI chatbot experience via an LLM.

Bimetric AI is a digital customer representative that provides 24/7 sales and marketing support for marketplace stores. It learns the content and products of brands and produces special answers to customer problems. Easily integrated for 3 minutes, Bimetric AI helps you increase your sales conversion by determining your most frequently asked products and trending topics with automatic or tools for full control.

Social AI

It is an interaction bot that can connect to live broadcasts on social media, follow broadcast streams, read comments, and provide feedback as comments. The AI robot can also work as a moderator.

Samet Product Information Mobile Application

We are developing a mobile application that will provide the sales team with fast and accurate access to detailed product information. The application extracts, categorizes and processes product data (from PDF documents). Thus, it will provide instant answers to complex product-related problems. Core functions include product comparisons, a natural language processing (NLP) based question-answer system and the ability to display rich answers in the form of text, images, links and data tables. The project is being prepared in multilingual formats, including English and Turkish.

Demand Forecasting

Project Overview

The goal is to forecast product demand using historical and external data.
Helps improve stock planning, reduce costs, and increase customer satisfaction.

Data Sources

Sales history, campaign data, holidays, and weather information.
Data collected on a daily and store-based level.

Approach

Time series forecasting techniques were applied.
Both traditional and AI-based models were tested and compared.
External factors like weather and promotions were included in the analysis.

Model Selection

Several models were evaluated to find the most accurate one.
The best-performing model was selected based on past prediction accuracy.

Evaluation

Forecasts were compared with actual sales.
Accuracy rate of over 80% was achieved.
Visual results were used to support interpretation.

Key Benefits

Prevents overstock and stockouts.
Supports decision-making for campaigns and supply planning.
Increases operational efficiency and customer satisfaction.

Challenges Faced

Inconsistent data and sudden demand spikes.
Large number of products and stores to manage.
These were addressed through data cleaning and grouping strategies.

Next Steps

Integration with internal systems for automated planning.
Continuous model improvement with new data and feedback.
Expansion to new product categories and stores.

Store Manager Assistant

Project Overview

A smart assistant (chatbot) was developed to support store managers in daily operations.
It provides quick access to store-specific data and answers to frequently asked questions.

Purpose

To save time for store managers by automating routine queries.
To help them make faster and more informed decisions.

Main Features

Answers questions about daily sales, stock status, returns, and performance.
Supports natural language queries (e.g., “What were my sales yesterday?”).
Can filter results by product, date, or store section.

How It Works

Connected to internal systems to fetch real-time and historical data.
Designed to understand simple, everyday language.
If a store name is not mentioned, it assumes the user’s own store.

Key Benefits

Reduces time spent on checking reports manually.
Increases operational efficiency and responsiveness.
Makes data more accessible and actionable for store staff.

User Feedback

Positive response from store managers regarding ease of use.
Helped reduce dependency on support teams for routine information.

Next Steps

Adding voice interaction and mobile access.
Expanding to cover HR, customer service, and logistics topics.

HR Assistant

Project Overview

A virtual HR assistant was developed to support employees and managers with quick access to HR information.
It simplifies HR processes by answering common questions and providing instant guidance.

Purpose

To reduce the workload on HR teams by automating repetitive queries.
To improve employee experience through faster and more transparent communication.

Main Features

Responds to questions about leave balances, payroll dates, shift schedules, and HR policies.
Guides employees on how to apply for leave, report an issue, or update personal information.
Personalized responses based on user role and department.

How It Works

Integrated with internal HR systems to access real-time data.
Understands natural language and adapts to company-specific terminology.
Ensures data privacy and access control by verifying user identity.

Key Benefits

Saves time for both employees and HR staff.
Improves accuracy and consistency in HR communications.
Available 24/7 for instant support across devices.

User Feedback

Employees appreciated the speed and simplicity.
HR teams reported a noticeable drop in routine inquiries.

Next Steps

Add support for voice interactions and mobile app integration.
Expand to cover training requests, onboarding, and performance feedback.

Finance Assistant

Project Overview

A financial assistant chatbot was developed to support finance teams with instant access to key financial data.
It helps users get answers to finance-related questions quickly and accurately.

Purpose

To improve decision-making by making financial data more accessible.
To reduce time spent searching through reports or asking colleagues.

Main Features

Answers questions about daily revenue, cost breakdowns, budget vs. actual comparisons, and financial KPIs.
Supports queries by store, product category, time period, and channel.
Provides visual summaries and formatted tables when needed.

How It Works

Integrated with financial databases and reporting systems.
Understands natural language queries (e.g., “What was last month’s revenue by region?”).
Defaults to a specific store or time range if not specified by the user.

Key Benefits

Speeds up access to critical financial insights.
Reduces dependency on analysts for basic reporting needs.
Enables better monitoring of financial performance across the organization.

User Feedback

Finance staff and managers found it intuitive and time-saving.
Helped reduce email traffic and manual report requests.

Next Steps

Integration with planning and forecasting tools.
Enabling alerts for unusual financial patterns or threshold breaches.

Customer Feedback Analysis

Project Overview

A sentiment analysis system was developed to analyze customer feedback from multiple channels.
The goal is to better understand customer satisfaction and identify common issues.

Purpose

To collect and analyze comments from platforms like Google Play, App Store, Şikayetvar, and internal systems.
To help departments take action based on real customer insights.

Main Features

Automatically classifies comments as positive, negative, or neutral.
Detects key topics such as delivery, pricing, in-store service, and product quality.
Provides percentage breakdowns for each topic and sentiment category.

How It Works

Uses AI to process large volumes of text and identify patterns.
Groups feedback by source, store, date, and subject.
Offers filtering and visualization features through a user-friendly dashboard.

Key Benefits

Helps identify the most common customer complaints and compliments.
Supports faster and more targeted responses to customer needs.
Enhances customer experience by turning feedback into actionable insights.

User Feedback

Widely used by marketing, operations, and customer experience teams.
Helped reduce response time to negative feedback and improved service planning.

Next Steps

Real-time monitoring and alerts for sudden spikes in complaints.
Expanding the system to analyze audio feedback from call centers.

Call Center Assistant

Project Overview

An AI-powered support system was developed to assist call center agents during and after customer interactions.
It helps improve service quality, reduce call handling time, and extract valuable insights from conversations.

Purpose

To automate conversation summaries, detect key topics, and analyze sentiment.
To support agents with relevant information in real time and improve post-call reporting.

Main Features

Converts customer-agent calls into text using speech-to-text technology.
Summarizes conversations and identifies main issues, customer sentiment, and action items.
Highlights frequently mentioned keywords and topics for analysis.

How It Works

Uses natural language processing to understand and analyze audio conversations.
Differentiates between the agent and the customer in the transcript.
Provides visual summaries and insights through a user interface.

Key Benefits

Speeds up post-call documentation and follow-up actions.
Helps managers monitor quality and detect recurring problems.
Enables faster training and coaching through real case insights.

User Feedback

Agents reported reduced workload and faster wrap-up times.
Supervisors gained better visibility into customer pain points.

Next Steps

Enable real-time suggestions during calls.
Integrate with CRM and ticketing systems for seamless action tracking.

Computer Vision Based Quality Control

Project Overview

A computer vision system was developed to perform automated quality control by analyzing product images.
It compares physical products with their digital designs to detect production defects.

Purpose

To minimize human error in visual inspections.
To ensure product consistency and reduce defect rates.

Main Features

Compares images of physical items with reference designs using AI.
Detects defects such as misalignment, missing parts, color differences, or shape inconsistencies.
Highlights faulty areas automatically for further review.

How It Works

Uses computer vision and deep learning to analyze images taken during or after production.
Flags deviations from expected standards based on predefined tolerance levels.
Allows operators to review and confirm the results through an interface.

Key Benefits

Increases inspection speed and accuracy.
Reduces costs associated with faulty products and returns.
Provides consistent, objective evaluations regardless of human fatigue or bias.

User Feedback

Quality control teams reported increased efficiency and confidence in inspection processes.
Helped reduce rework and waste in production.

Next Steps

Integrate with production lines for real-time inspection.
Expand the model to support multiple product types and lighting conditions.

Computer Vision-Based Ripeness Level and Potential Disease Detection of Agricultural Products

Project Overview

An AI-based vision system was developed to detect the ripeness level and potential diseases of agricultural products using camera images.
It helps producers monitor crop health and optimize harvest timing.

Purpose

To provide early detection of plant diseases and improve harvest quality.
To reduce manual labor and human subjectivity in assessing crop conditions.

Main Features

Classifies fruits or vegetables based on ripeness levels (e.g., unripe, ready, overripe).
Identifies visual symptoms of diseases like discoloration, mold, or surface damage.
Provides alerts and visual indicators through a dashboard.

How It Works

Uses cameras placed in the field or greenhouse to capture real-time images.
Applies deep learning models trained on labeled image datasets.
Analyzes color, texture, and shape to determine maturity and health status.

Key Benefits

Supports timely harvesting decisions, leading to better product quality.
Helps prevent the spread of disease through early intervention.
Reduces dependency on expert field inspections.

User Feedback

Farmers and agronomists found it useful for daily field monitoring.
Improved planning for harvest and disease treatment.

Next Steps

Integrate with drone and mobile camera systems for large-scale monitoring.
Add prediction capabilities for disease risk based on environmental data.

Big Data Platform (Cassandra)

Project Overview

A robust Big Data platform was developed to manage and analyze large-scale data from various sources within the organization.
The platform is designed to handle complex data processing, storage, and real-time analytics.

Purpose

To enable data-driven decision-making by providing fast and reliable access to large datasets.
To support advanced analytics and machine learning applications at scale.

Main Features

Centralized data storage and processing across multiple sources (e.g., sales, inventory, customer feedback).
Real-time data ingestion and batch processing capabilities.
Seamless integration with analytical tools and AI/ML models.

How It Works

Uses Hadoop and other distributed systems for efficient storage and parallel data processing.
Supports ETL (Extract, Transform, Load) pipelines to clean, format, and organize raw data.
Scalable infrastructure to accommodate growing data volumes.

Key Benefits

Enables faster insights and reporting across departments.
Reduces data silos and promotes cross-functional collaboration.
Provides a scalable foundation for future growth and data expansion.

User Feedback

Teams reported faster data retrieval times and more efficient data handling.
Helped improve decision-making with real-time access to relevant data.

Next Steps

Implement advanced analytics and AI capabilities to predict trends and patterns.
Expand the platform to include additional data sources and improve data visualization.