AIRA — A worthy proponent of CapFlow

• This post is part of a series of blogs on Capillary Technologies’ Artificial Intelligence (AI) capabilities. The first blog talked about CapFlow — Capillary’s implementation of Machine Learning Operations (MLOps). In this post, we discuss Artificial Intelligence Retail Analytics (AIRA) which is the product built on top of CapFlow engine to empower AI capabilities.

Computer systems generate vast volumes of data and we have efficient techniques to capture data. When data is used to gather insights, there could be potential productivity gain for an enterprise.

Raw data must be engineered such that we can build mathematical models and forecast profitable or untoward circumstances. Since the importance of data was identified relatively recently, we need to have a facility to retrofit data processing systems into established systems.

MLOps define a set of best practices that aim to efficiently deliver data science solutions to enterprises. Capillary Technologies’ own framework CapFlow is built on the solid principles of MLOps. In this post, let’s discuss AIRA, a worthy offshoot of CapFlow, and gain an insight into how AIRA redefines Capillary’s data science endeavor.

The need for data engineering in the Capillary ecosystem

Capillary Technologies is an enterprise whose products generate and leverage massive amounts of data. For instance, when Capillary’s products are deployed worldwide, customer transaction numbers could range in billions. Likewise, new customer registrations are legion.

Capillary Technologies is an ideal example of how raw data could be harnessed to derive inference thereby aiding organizations to execute better campaigns. AIRA dashboard works hand-in-hand with CapFlow and features personalization with just a few user clicks.

A bird’s eye view of the AIRA dashboard

AIRA dashboard is a web application that abstracts the complexity of data science and provides an intuitive interface with minimum user interactions. Although user intervention is scarce, AIRA does not compromise on effectiveness and features cutting-edge models.

There are several measurable pieces of data generated by Capillary products that could be identified as features. Such features are effectively quantified by CapFlow and a model is constructed along with other features.

At the outset, AIRA features various AI-ML models including transaction prediction, campaign response prediction, and customer churn prediction. It is quite evident from the names that transactions, campaign response, and customer churn are salient parameters warranting a deep inquiry.

AIRA dashboard features upcoming models and thus mitigates the drudgery to build, train or deploy models. For instance, campaign time affinity, offer affinity, and seasonal affinity are some models identified to play a vital role for organizations to scale their customer outreach.

AIRA dashboard — featuring personalization with bare-bones input

Software interfaces must be simple to be effective. Too many input elements vitiate the look and feel of an application. AIRA dashboard features a simple interface and any personnel with relatively no background in data engineering would be able to go hands-on in a short time.

When a user goes live on the AIRA dashboard portal, they are presented with an intuitive real-time view that conveys the models enabled and their current status. The user can go ahead and enable new models. The conventional facilities to filter models based on name or their status are readily provided.

AIRA dashboard streamlines the process of data validation and training which makes it unnecessary for a user to get down to a granular level. If there is an error during the validation phase, the incident is flagged and reported to the user.

The highlight of the AIRA dashboard is the way in which it guides users from phase to phase with clear messages. AIRA dashboard explains the purpose of each model and provides an insight into each phase which makes it easy to breeze through the product.

The conclusive effectiveness of data science models is analyzed with various evaluation metrics. AIRA dashboard offers the best set of metrics like accuracy, precision, recall, F1 score, mean ROC, and hit rate. These metrics offer an insight into how the model is faring and whether it needs attention.

AIRA dashboard makes it convenient to relay the results of data modeling to Capillary’s products like Insights+ and Engage+. Ultimately, AIRA dashboard has minimized user clicks and enables the user to use data from the perspective of an analyst.

Conclusion

AIRA dashboard and CapFlow are cornerstone applications that are well poised to lead Capillary’s data science endeavor to success. Artificial intelligence and machine learning are paradigms that are in vogue and Capillary is set to achieve greater heights backed by its ventures!

The think tank behind the AIRA dashboard includes Biswa Singh, Saurav Behera, Anuja Kolse, Bikash Bhoi, Rishabh Ojha, Santosh Kumar, Akshath Varugeese, Arpan Malik, Arpit Garg, and Harundas Koonhavil from the technical team. From the product team, the contributors were Jyotiska Bhattacharjee, Anjaney Vatsal, Shaan Shivanandan, Sarth Talati, Sunil Dhananjaya, and Achyu.

Upcoming Features

• Alerts: A notification system, where users can select at what stages of the model they want to be notified.
• Better accuracy report: An accuracy report with all the monitoring parameters depicted in the form of bar charts or graphs.
• Data validation report: A report generated after the validation stage is completed.
• Deeper integration with Capillary products (Member Care, Engage+ and Insights+)