We combine our expertise in Digital Signal Processing (DSP), Big-Data and Artificial Intelligence (AI) to provide unique problem-solving expertise. Our team has been involved in product development for nearly hundred products in these two ways:

Custom Research & Development

In this scenario, we research or develop for your specific use case. For instance, we have developed customized and highly-performing DSP algorithms in C and Python for highly critical systems in the healthcare sector. We implemented tailored LLM models for specific use cases in the retail and supply chain industry. Custom development may or may not use our intellectual property (IP).

Licensing

For almost 10 years, we have accumulated know-how and developed problem-solving frameworks and optimized code. We have dozens of ready-to-use technologies. In addition, we can generate product-ready libraries from our Digital Signal Processing and Machine Learning libraries.

Let’s take a look at a couple of use cases where we engaged with our clients taking DSP and AI solutions into production using our problem-solving framework.

A medical device startup headquartered in Mountain View in the San Francisco Bay Area is launching the next generation of wearable devices capable of analyzing vital signs from patients. They need to launch a product that can process time-series data coming from several sensors embedded in their device and extract relevant insights helping patients manage their health in real-time. The data coming from the sensors include temperature, acceleration, electrocardiography (ECG) and photoplethysmogram (PPG) signals. The final algorithm needs to be developed following FDA guidelines.

We have experience developing complex and highly-critical algorithms for the healthcare industry under strict FDA standards. We use DSP to extract time and frequency domain features for each data stream coming from the sensors using state-of-the-art algorithms. We use those features to create ML models and predict patient signals such as heart rate, oxygenation level and blood pressure. We use our problem-solving framework to prioritize, integrate and deploy this solution into the client infrastructure consisting of a wearable device, a smartphone application and a cloud infrastructure. The deployment is done following best software practices for future FDA submission.

Upon completion of the project, the client raises $35M in additional funding which led to a successful exit in the form of an acquisition.

A major U.S. RFID manufacturer and IoT platform software provider is seeking to increase their market share in the retail market. They want to differentiate themselves by offering their clients a comprehensive solution allowing them to track real-time store information, predict customer behavior and increase conversion rates allowing their clients to increase profit.

The client wants to analyze and extract features from their time-series data streaming from millions of RFID sensors. The data contains device, product location and store layout information. Our team uses DSP techniques to combine these features with customer transaction data which is then fed into an initial ML model. After several iterations, the models allow to predict indicators such as conversion rates, common customer paths and product movement patterns. Our team deploys these models in production by porting them into Apache Spark which are then deployed into a third-party cloud service and connected to an IoT ingestion engine. Finally, a big-data visualization platform processes the output and offers our client's customer the ability to optimize store layout and reallocate resources.

A San Francisco wellness startup proposes a system that measures blood chemistry at home and delivers personalized and actionable insights. The company wants to leverage their hardware technology to detect key blood chemistries (including glucose, triglycerides, LDL, HDL, and total cholesterol level) indicative of cardiovascular health. 

Using their blood sampling device, large amounts of data are being collected in the form of multivariate spectrophotometry data within and between patients. This enables longitudinal sampling across various conditions, demographic, and geographic groups.  In addition, to deriving temporal trends and forecasting from the user’s own data, the system can monitor and refine its health recommendations. The system continuously adjusts its learning models through the application of reinforced learning mechanisms.

We propose a dedicated and systematic effort to implement a machine learning pipeline using dedicated inference, distributed data storage and parallel processing. The initial focus is to design and develop the DSP extraction routines to obtain features from time-series data. Next, the focus switches to the development of the ML prediction algorithm to estimate basic blood chemistries using medical and genetic profiles. Finally, the pipeline is deployed in a computing cluster using Apache Spark. In addition, verification and validation procedures are executed to comply with FDA regulatory guidelines.