Artificial Intelligence - Clustrex Data

Artificial Intelligence Services

At Clustrex, we deliver state-of-the-art AI and machine learning solutions designed to transform business operations and customer experiences. Our offerings include advanced chatbot systems, image processing tools, document management solutions, and other innovative applications tailored to meet diverse industry needs.

Our chatbots are expertly developed to enhance customer engagement, automate workflows, and facilitate seamless data collection through natural, conversational interfaces. They integrate seamlessly with client websites to interact with users, gather insights, and query databases for dynamic dashboards and actionable reports.
In image processing, our solutions address a broad spectrum of applications, from facial recognition and virtual try-ons to groundbreaking healthcare innovations like disease detection from scanned images.
Our document management tools further drive efficiency by automating tasks such as data extraction, summarization, and information processing, ensuring greater accuracy and productivity.

Discover how Clustrex's AI expertise can empower your organization with intelligent solutions that deliver real-world impact.

1. Dynamic Project Query Bot

User-Friendly Queries :
The bot allows users to interact intuitively with project data through natural language queries.
Dynamic Information Retrieval :
Users can obtain specific details, including open and completed project lists or in-depth information about individual projects.
Sorting Capabilities :
The bot supports sorting projects by completion dates, providing a quick overview of project timelines
Status-based Filtering :
Users can easily filter projects by status, streamlining access to ongoing, completed, or open projects.
Value Inquiry :
The bot accommodates queries about project values, aiding in financial planning and decision-making.
Deadline Proximity Alerts :
Users can inquire about projects ending soon, facilitating prioritization and efficient project management.
Top-Value Sorting :
Users can request a sorted list of completed projects based on values, supporting analysis of high-value projects.
Intelligent Responses :
The bot is programmed to comprehend diverse queries, ensuring accurate and relevant responses.
Enhanced Experience :
By incorporating these features, the bot aims to elevate the overall user experience, promoting effective project management and decision-making.

Try Out

AI based Image processing

Clustrex provides cutting-edge AI and machine learning solutions tailored for diverse image processing applications. From facial recognition for personalized services to advanced deep learning techniques, we deliver intelligent systems that enhance accuracy and efficiency across industries.

Our expertise includes:

Facial Recognition: Streamlining security with facial authentication and providing detailed insights for personalized services.
Virtual Try-Ons: Revolutionizing e-commerce and eyewear with seamless, AI-driven virtual try-on technology.
Advanced Image Analysis: Utilizing deep learning models like YOLO and AWS Rekognition to transform raw image data into actionable insights.

At Clustrex, we harness the power of technologies like deep learning to unlock limitless potential. Let us empower your business with innovative AI solutions designed to drive growth and transform operations.

Face Detection App

Experience the future of facial recognition with our AI-powered tool. Introducing our cutting-edge facial recognition tool, a versatile solution designed for seamless integration into various platforms. With the ability to recognize faces, this tool is perfect for streamlining processes such as secure check-ins, efficient attendance tracking, and personalized user experiences. Tailored to your needs, our tool can be customized for tablets, computers, or CCTV cameras, providing a reliable and adaptable solution for diverse applications. Elevate your systems with the power of facial recognition technology.

Glass Tryon

Introducing our revolutionary AI-powered virtual try-on tool! Say goodbye to the hassle of visiting multiple stores to find the perfect pair of eyeglasses. With just a webcam or laptop cam, our advanced facial recognition technology seamlessly integrates with vast collection of eyewear frames. Explore a wide range of styles and see exactly how they look on your face in real-time. If you are a retailer looking to expand your offerings, our tool revolutionizes the eyewear shopping experience. Try it today and discover your perfect pair effortlessly!

1. Disease Classification in OCT Scans Using ResNet-50

Objective :
The study aims to sort retinal images into four categories: Choroidal Neovascularization (CNV), Diabetic Macular Edema (DME), Drusen, and Normal. This helps in detecting eye diseases early on.
Dataset :
This dataset uses retinal OCT images (like an eye scan) labelled CNV, DME, Drusen, or Normal, with thousands of pictures in each category.
Model Structure :
It uses ResNet-50, a deep learning model designed to improve accuracy and has “residual blocks” that help it learn well, allowing it to identify complex patterns.
Transfer Learning :
The model is adjusted to work with eye images specifically by replacing its last layer to make it capable of identifying the four eye categories.
Training Setup :
The dataset is split, with 80% used for training and 20% for validation. The model uses cross-entropy loss and the Adam optimizer, with early stopping to make training more efficient.
Evaluation Metrics :
The model’s performance is checked based on accuracy, precision, recall, and F1 score. A confusion matrix is used to see which images it misclassified.
Results & Conclusion :
The model achieves 97% accuracy on test images. The model effectively identifies eye diseases, showing potential for use in clinics to improve early detection and patient care.

2. Implementation of AWS Rekognition for Patient and Staff Tracking in a Clinic

Introduction :
AWS Rekognition tracks patient and staff movements, analyzing metrics like time spent, wait times, and staff interactions to improve clinic operations.
Technologies Used :
Python, AWS Rekognition, ECR, S3, Lambda, EC2, SNS, PostgreSQL, and YOLO.
Live Feed Data Collection :
Cameras in strategic locations track individuals' presence and time in the clinic.
Multiprocessing Using Python :
25 cameras capture 0.5-second clips every 5 seconds. Python’s multiprocessing optimizes AWS Rekognition face detection and tracking.
Local Server with GPU :
A GPU-powered server uses YOLO to detect human presence in clips, sending only relevant footage to AWS Rekognition.
YOLO11S :
YOLO11S enhances real-time object detection, ideal for dynamic clinic environments.
Good Quality Images for Indexing with AWS Rekognition :
High-quality images are indexed in AWS Rekognition collections for fast identification of patients and staff.
Advantages of Using Video Input for Rekognition :
Continuous detection, multiple recognition opportunities, and enhanced accuracy improve patient and staff tracking.

3. Color Filtering in Images

Objective:
Create a method to spot, analyze, and measure red-coloured areas in OCT images of the retina that suggest Geographic Atrophy (GA). This technique enhances early detection, allowing clinicians to track disease progression more effectively and implement timely interventions.
Problem Statement:
Accurately finding and measuring red areas in OCT images is essential for understanding the extent and severity of GA.
Goals :
The method aims to (1) detect red areas in OCT images that indicate GA, (2) analyze the shape and distribution of these areas, (3) measure the affected regions to assess disease severity.
Methodology:
The images are converted to make it easier to identify red areas. Contour and edge detection are used to mark the boundaries, showing the shape and placement of affected regions. These outlines are then overlaid on the original images for easier viewing. To assess GA severity, the red areas are measured by counting pixels and converting this count to an area size in mm². The entire process is automated to analyze multiple images consistently and efficiently.
Results:
The outputs include images with clear boundary overlays, highlighting affected areas, and data files with pixel counts and area measurements in mm². This allows for easy comparison of affected regions between patients and over time.
Conclusion:
This process provides an accurate, automated way to detect and measure GA-affected regions, helping doctors with patient assessments and supporting better-informed care decisions.

4. Smart AI Solutions for Virtual Try-On Applications Preparation

Objective :
Create a model based on U-Net to automatically segment eyewear frames and lenses in images, simplifying virtual try-on preparation and reducing manual editing.
Dataset :
The dataset includes 200 carefully selected images with labelled masks for frames and lenses. Images are split 80% for training and 20% for testing.
Model Structure :
The model uses U-Net, which has two main paths: one for extracting features and another for refining the details, making it effective for precise segmentation.
Training:
The model is trained on a GPU, with a learning rate of 0.001, and a scheduler to adjust learning. Key challenges include accurately capturing the edges of transparent lenses and thin frames
Evaluation:
The model’s performance is measured using a metric called Binary Cross-Entropy (BCE) Loss, achieving a test loss of 0.04.
Deployment:
The model is set up on a server with GPU support for the efficient processing of images.
Limitations:
Challenges remain with images where the frame, lens, or background share similar colours, and with handling lens transparency and reflections.
Conclusion:
The model shows strong results in standard conditions and provides a good foundation for future improvements, such as training with more diverse data and refining techniques to handle tricky cases.

5. Virtual Try-On System for Opticals and E-Commerce

Introduction :
AR and AI are revolutionizing online eyewear shopping, allowing customers to visualize frames on their faces. Our technology ensures highly realistic, accurate, and real-time virtual try-ons, mimicking the feel of physically trying frames in-store.
Problem :
Online shoppers often feel uncertain without proper visualization, leading to higher return rates. We aim to replicate the in-store try-on experience online.
Solution :
By using Facial Landmark Detection, our system precisely positions frames on users' faces. The glasses fit naturally with facial features, adjusting to each person’s unique shape, and enhancing their confidence when exploring different styles.
Benefits:
It provides a realistic try-on experience, boosting user satisfaction. Its cross-platform compatibility ensures smooth performance on both mobile and web, providing a versatile and accessible user experience.
Challenges :
Limitations include sensitivity to lighting conditions and lower performance on less powerful devices.
Conclusion :
Our virtual try-on solution transforms online eyewear shopping by offering a realistic, in-store-like experience. This advanced AR tool helps reduce return rates and enhances the decision-making process, giving retailers a scalable way to elevate customer satisfaction.

6. Personal Protective Equipment Detection Using YOLO

Overview :
Ensuring PPE compliance (helmets, goggles) is crucial for workplace safety. Manual monitoring is time-consuming and error-prone. Using YOLO, a real-time object detection model, enables efficient detection of PPE violations.
Objective:
Develop a system for real-time PPE detection to improve safety and compliance.
Methodology:
A dataset was curated with images of workers both with and without PPE under varied conditions, annotated with bounding boxes. The model, trained with transfer learning and data augmentation, achieved 92% accuracy for helmets and 84% for goggles, at 20-30 FPS on Nvidia GPUs. Testing included metrics like mean average precision (mAP) and a confusion matrix for precision checks.
Deployment:
Deployed on Nvidia GPU-equipped edge devices. Cameras in high-traffic areas trigger alerts for PPE violations.
Challenges:
Issues like occlusion, lighting changes, and complex worker postures caused some misclassifications, which were mitigated through dataset augmentation and fine-tuning.
Results:
Improved compliance, fewer manual checks, and enhanced safety audits.
Future Scope:
Expanding to drone-based monitoring for broader areas and incorporating advanced analytics to track compliance trends and predict potential issues.

Document processing with AI

Clustrex offers advanced AI-driven solutions that redefine document processing, enabling businesses to streamline workflows, enhance accuracy, and save time. Our tools are designed to tackle a wide range of document types, from invoices and bank statements to resumes and contracts. Here's how we transform your document management processes:

Data Extraction: Effortlessly extract structured data from PDFs such as invoices, medical forms, and identity cards. Our precision-trained AI provides outputs in CSV, Excel, and JSON formats, tailored to your specific needs. Simplify data handling with unmatched accuracy and ease.
Named Entity Recognition (NER): Extract and structure critical information like names, qualifications, and work experience from resumes and documents. Ideal for HR departments and recruiters, this tool boosts productivity and decision-making, enabling efficient talent evaluation and data management.
Retrieval-Augmented Generation: Go beyond data extraction with our versatile tool. Upload any document—contracts, receipts, or reports—and let our AI retrieve relevant information, calculate totals, generate summaries, or answer specific queries. Say goodbye to manual data entry and embrace efficiency with AI-powered insights.

Experience the future of document processing with Clustrex's innovative AI solutions, designed to empower businesses and unlock new possibilities.

Case Study

1. Bank Statements Extractor

Clustrex has developed a robust web application designed to streamline the process of handling bank statements. Our solution allows users to effortlessly upload statements from various banks, automating the extraction of data into a standardized Excel format. This eliminates the need for manual data entry, saving valuable time and reducing the risk of human error.
Users have the flexibility to create templates tailored to each bank's requirements. Alternatively, they can utilize built-in templates that encompass all necessary fields, providing adaptability to diverse user needs.
We have also integrated automatic verification system, which reduces user's time to manually verify all the files. This ensures the accuracy and integrity of the extracted data.
Technologies Used :
AWS Textract, S3, Lambda, DynamoDB, API Gateway, Google Login.
Value delivered :
We have also integrated automatic verification system, which reduces user's time to manually verify all the files. This ensures the accuracy and integrity of the extracted data.
Challenges :
Converting variety of formats used by each bank to a single standard format.

2. Blog Generator for Bankruptcy documents

A web application to create blog based on the documents uploaded by the user. The application is integrated with an AI powered chatbot for content generation. The content is intelligently extracted and transformed into a cohesive and engaging blog format.
Enhanced user interaction with our AI-powered chatbot, designed to provide customized results based on user prompts. This feature ensures that users can seamlessly incorporate tailored information into their blogs.
Technologies Used :
AWS Bedrock, Langchain, AWS Textract, S3, Lambda, DynamoDB, API Gateway, Google Login, Prompt Engineering. LLM modals: GPT-4 from OpenAI, Claude from Anthropic.
Value delivered :
Application not only acts as blog generator, but also acts as a chat-bot and answer user's queries based on the documents provided by the user. By adopting a serverless architecture, we have optimized costs while maintaining high efficiency and scalability.
Challenges :
The complexity of transforming document data into informative blogs and integrating advanced AI models.

3. Invoice Extraction from a variety of vendors

Each business organization has their own invoice template and is a very difficult task to manually get the data from each invoices and organize them in a structured format for payments and other uses when you are an organization with large supplier base. At Clustrex, we have created an application that processes files once uploaded into a S3 bucket and extracts the data into a structured format.
The platform empowers reviewers to verify the extracted data and also facilitates easy uploading of files. Integrated verification mechanisms, including confidence score calculations, guide reviewers in assessing the accuracy of extracted data, streamlining the review process.
We have embedded GPT-4 powered LLM to extract LineItem fields from the file, even when presented in multiple tables. The extracted data is consolidated into a single structured table, enhancing readability and usability.
Technologies used : AWS Bedrock, OpenAI's GPT-4, Prompt Engineering, Llama-Index, AWS Textract, S3, Lambda, Comprehend, Sagemaker Groundtruth, EC2, Step Functions, SQS, MySQL. LLM modals: GPT-4 from OpenAI, Claude from Anthropic.
Value Delivered : High Quality Data, ability to easily review only the required files, reduced manual intervention for files.
Challenges : To achieve a standard format for various vendors and to ensure consistency and reliability in handling diverse invoice templates.

Clustrex Data Private Limited

Say Hello!

Email

Phone

Address

Madipakkam Office

Request A Demo