Customer Analytics
(Retail/Consumer)
Problem
Identify happy and unhappy customers in real-time and take corrective measures to hold back the customer
Solution
Customer Analytics
(Surveillance)
Problem
Analyze continuous video feed, from remote sites. Each site had 4 IP cameras connected to a DVR, recording video feeds 24/7. The cameras had to automatically identify and raise appropriate alarms. For example, if any object went missing or there was an intruder within their perimeter the requirement was to notify the person(s) in-charge of the site.
Solution
Gait Analyzer
(Healthcare)
Problem
To find the Gait of a patient without any contact mechanism and in a short time.
Solution
We used non-invasive, contactless methodology to assess the Gait of a patient. Taking advantage of the depth sensing and motion sensing technology, we used a depth sensor to measure the lower body parameters. We were able to gather the real-time Gait data of the patient walking in front of the depth sensor and plot it on the screen for the doctor to monitor. The software then uses the prediction model to recommend what next physiotherapy the patient should undergo.
Tools: : Installable Linux application built using Python.
Live Feed Analytics
(IoT)
Problem
Solution
The client had various IoT devices in different sites, gathering real-time data. These real-time data were pooled and stored on various servers. They had a problem of managing, visualizing, and analyzing the large volume of data.
We developed a Hadoop based platform for data handling. We configured an application on Kubernetes, for real-time data streaming and analytics. The JS based visualization dashboard had various charts and graphs for plotting the data in real-time. The tool included a fully automated rule engine to customize triggers. We used machine learning algorithms to predict the data we were receiving. We developed a predictive maintenance engine for the machineries. Tools: : Hadoop, Python, MEANStack framework.
Crypto Trading / Finance
(DeFi)
Problem
Solution
There are 100s of crypto currencies that are being traded in 1000s of exchanges across the world. Unlike regular trading, cryptos are vulnerable for price manipulation. There are numerous spoof trades that are placed on the exchanges which go unnoticed. This can significantly impact real traders. Our objective: Identify these spoof trades in real-time.
We took the historical data of various cryptos from various exchanges, and generated forward- and backward-looking signals for identifying markers for spoof. We applied this over a larger dataset from multiple exchanges. Then we refined the predictive algorithm so we can use the signals as features to predict spoof activity. Tools: : MathLab, Python, D3Js for Visualizations
Electrochemical Data Analysis
(Defense)
Problem
Solution
We had to identify the physical characteristics of electrical signals. The sensor identifies various gases and converts them to a series of electrical signals. We had to develop an algorithm that: (1) detects the peak from individual signals, (2) identifies concentration from unknown pure signals, and (3) finds the resolution of a mixture signal into individual components.
We used the following approach to arrive at a solution. (1) To detect the peak from individual signals: We employed a heuristic-based algorithm to detect the peaks of the individual signals. (2) To identify concentration from unknown pure signals: We developed a machine learning algorithm that learned the relationship between peak area and peak height with actual concentrations. (3) To find the resolution of a mixture signal into individual components: We assumed a quasi-linear relationship on how the individual signals behaved in a mixture, given the closest possible individual components to the mixture, the machine learning algorithm learns a quasi-linear relationship to solve this problem. Tools: : Python.
