Our services and core competencies

Machine learning

Supervised and unsupervised learning, neural nets, graphical models, mixture models and EM, HMM's, time series analysis.

Computer vision

Object recognition and detection, pose recognition, character recognition, video processing.

Image processing

Visual signal detection, seamless cloning, image inpainting and visual pattern detection.

Data pipelines

Cassandra, PostgreSQL, data processing, storage and latency optimization, web services, API

Data science

Numpy, Scipy, data analysis and insights, data cleaning, modeling and visualisation.

Software development

Python, Java, C++, C# .NET, API, React, JS, Cloud, mobile and desktop applications.

Our work
Demand forecasting

We have developed an automated sale forecasting system for e-commerce stores which uses historical data to learn demand patterns. Our solution involves both unsupervised and supervised learning for data modeling. The work encompassed developing scalable backend and databases for storing historical data, automated data pipelines for daily model training and supporting responsive dashboard and API's.
The work includes extensive model quantification and algorithm engineering, automated predictive API and scalable data infrastructure.

Dynamic pricing

We have developed automated dynamic pricing solution which optimizes product prices in an e-commerce store to achieve a chosen business goal. Prices can be optimised based on custom pricing strategies. Historical store data with pageviews are used to model conversions using supervised learning. Models are trained daily and regular price split tests can be deployed to quantify benefits of chosen pricing strategy.

Pose recognition in video

We have developed an automated pose recognition system, which analyzes video of a runner recorder with iPhone camera and provides automatic analysis of runner's technique. The product enables people run faster, reduce injuries and improve running style. Automatic pose analysis is performed using computer vision, image processing and supervised machine learning. Our work involved all standard steps necessary for pose recognition including training data set preparation, algorithm research, optimization and training and intensive time and memory tuning.

pose recognition in video
Real-time on device optical character recognition

We've engineered and trained a real-time OCR and text classifier for augmenting words with information from Wikipedia and dictionary. This project involved intensive accuracy and speed optimisation under low-memory footprint on iPhones. The work encompassed accuracy quantification, training, algorithm engineering and word segmentation. Challenges included blurred video stream and runtime environment with constrained memory.

real-time OCR
Image and video processing

We have delivered commercial projects and proof-of-concept works involving visual signal detection, seamless cloning, image inpainting and visual pattern detection. We have developed standalone libraries or our code has been directly compiled in an end-product. Our experience spans developing image processing libraries which are cross-platform and can run on desktop, server and mobile devices.