Discover our work

Software solutions we have delivered. We work hard and we are dedicated to producing the best possible outcome for our clients, as you can see from our case studies.

Website Builder Co-Pilot

Client: A UK-based no-code website builder platform

We built a robust AI-driven engine to automate the website creation process. Our work focused on training models for a virtual assistant that collects key input from users such as design preferences and business goals and translating that into automatically generated website layouts. The solution uses smart algorithms and predefined components to assemble fully functional, tailored web structures without manual design work.

Technologies: GPT-4, Elasticsearch, OpenCV, PyTorch, Node.js, GraphQL, FastAPI, Webflow API, Python,

3D Animation Generator for Games, AR & VR

Client: Game development company

We developed an AI-powered solution to automate the creation of 3D object and character animations for use in gaming, AR, and VR environments. Our system enables the rapid generation of natural-looking animations including character-object interactions-using a scalable engine that supports a wide variety of virtual assets such as vehicles, avatars, and dynamic props. By minimizing the need for manual rigging and keyframe work, the platform significantly reduces production costs and time-to-market while allowing broad flexibility in customizing asset behavior and animation style.

Technologies: PyTorch3D, Unity API, Blender Python API, OpenCV, FastAPI, Python, custom deep learning models for motion synthesis

Client: Retail company

We developed an AI-based tool for automated data extraction from unstructured documents. The solution converts diverse inputs such as invoices, receipts, and internal forms into structured key-value formats, enabling seamless transformation of raw files into business-ready data optimized for processing, analytics, and archival. Leveraging NLP and computer vision techniques, the tool identifies and maps relevant information with high accuracy, drastically reducing manual effort and ensuring data consistency across systems.

Technologies: GPT-4, Tesseract OCR, spaCy, PyTorch, FastAPI, Pandas, Python

AI-Powered Document Structuring Engine

Ad Spend Optimization Engine

Client: A US-based digital marketing intelligence platform

We built a machine learning system to optimize multi-channel advertising budgets by predicting the most effective timing and placement for ad campaigns. Leveraging historical performance data, the model forecasts engagement metrics—such as views, clicks, and conversions—and dynamically allocates spending to maximize return on investment (ROI). The solution enables marketers to make data-driven decisions, reduce wasted spend, and continuously refine campaign strategies through automated insights.

Technologies: XGBoost, LightGBM, Pandas, Scikit-learn, Python, Apache Airflow, PostgreSQL, AWS S3, Tableau, FastAPI

Retail Traffic Analytics Platform

In-house project

We developed a computer vision-based visitor analytics solution for physical retail environments, enabling precise measurement and interpretation of in-store foot traffic. The system leverages AI models to transform live camera feeds into actionable insights, supporting space optimization and customer experience enhancement. Key functionalities include people counting, visitor path tracking, heatmap generation (crowd density, dwell time, traffic flow), queue monitoring, and MAG (mood, age, gender) profiling. The tool empowers retailers to make data-driven decisions on layout adjustments, staffing, and promotional strategies.

Technologies: OpenCV, YOLOv3, TensorFlow, Python, FastAPI, PostgreSQL, Redis, React,

AI-Powered Attention Heatmap Engine

Client: Perceptbox – a visual content analytics startup

We developed an AI-driven simulation tool that predicts human attention distribution in the first moments of visual exposure. The platform generates attention heatmaps using advanced neural networks, accurately modeling how users perceive and prioritize elements in images, advertisements, and UI layouts. By mimicking the natural behavior of the human visual system, the solution enables designers and marketers to identify focal points, detect overlooked areas, and optimize content for maximum engagement. The system delivers rapid, automated insight without the need for manual eye-tracking studies.

Technologies: C++, OpenCV, Custom CNNs, Python, ONNX, Qt

Social Distancing Analytics System

Client: Public safety and urban analytics company

We developed an AI-powered system designed to monitor and evaluate social distancing in public spaces. The solution uses advanced computer vision models to detect individuals in real time and measure interpersonal distances. A visual indicator, based on a traffic light metaphor, highlights safe behavior (green/yellow) and violations (red) without compromising privacy. The system helps city authorities and venue operators pinpoint spatial bottlenecks, optimize crowd management, and track behavior trends as public health policies evolve.

Technologies: YOLOv7, OpenCV, Python, PyTorch, TensorRT, Flask, NVIDIA Jetson, PostgreSQL, WebSocket

Facial Attribute Recognition Engine

Client: A US-based biometric security platform

We designed and deployed a deep learning-based system for real-time gender, age, and emotion detection from facial imagery. The solution leverages curated datasets such as IMDB-WIKI and UTKFace, enabling robust classification even in varied lighting and occlusion conditions. It also includes an advanced eye segmentation module, which supports additional use cases like iris and eye color recognition. The system enhances facial authentication workflows, enables adaptive content personalization, and strengthens customer profiling in digital environments.

Technologies: TensorFlow, OpenCV, Keras, Python, ONNX, NVIDIA CUDA, Flask, FastAPI, Docker, PostgreSQL

Hair Segmentation Engine

In-house project.

We developed a deep learning–based solution for high-precision hair segmentation tailored to applications in virtual styling, augmented reality, and digital image enhancement. The engine utilizes a custom PyTorch model based on the U-Net architecture with a MobileNet v2 backbone, trained on the Figaro1k dataset to deliver pixel-perfect segmentation across diverse hair types and backgrounds. Designed for real-time performance, the lightweight model is optimized for deployment on both mobile and web platforms, enabling seamless integration into beauty and fashion applications and enhancing user engagement through personalized, AR-driven experiences.

Technologies: PyTorch, U-Net, MobileNet v2, Figaro1k dataset, ONNX, FastAPI, Python, OpenCV

Weapon Detection System (AiPOD 2)

Client: A security technology provider specializing in public safety infrastructure

We developed a real-time weapon detection system powered by computer vision and deep learning to enhance threat identification capabilities in sensitive environments such as transportation hubs, schools, and public venues. The solution is based on a custom-trained YOLOv3 model and is capable of detecting multiple weapon types including AK47, AR15, M4, Glock900, and SIG Sauer. Trained on a proprietary dataset of 10,000 annotated images, the system ensures rapid and accurate identification with low false positive rates. The lightweight architecture enables integration with CCTV and drone surveillance systems, supporting proactive threat mitigation and emergency response automation.

Technologies: YOLOv3, OpenCV, Python, CUDA, Flask, TensorRT

Back to Home Page