Omkar Shirhatti, CEO & Co-Founder, Karza Technologies, has developed a KYC solution for the BFSI sector. He discusses about it in this interaction.
How did you think about developing this solution?
The concept of a product that allows remote KYC was being discussed in the company for a few months. It fit in well with Karza’s vision of creating digital due-diligence products to empower financial institutions. So once RBI released Video KYC guidelines, we immediately set the ball rolling. There were some challenges we faced initially. While we envisioned the process to be fully automated, RBI mandated it to be agent assisted. This posed an additional challenge for our clients that would have to train their agents to use the video KYC application.
So while initially designing our system, apart from adhering strictly to the guidelines, we laid emphasis on making the process seamless and easy for not only customers but also agents. This resulted in our solution being easily adopted by financial institutions. While Karza product is now complete and being used by several companies, we’re constantly striving towards perfecting our product for clients – minimising their issues while maximizing throughput.
Does it use any emerging tech and analytics etc? Discuss in detail
A lot of exciting technology and analytics has been used in building Video KYC. Serverless components and fully-containerised distributed deployments (assisted with data-driven auto-scaling policies) enables scale-on-the-fly, cost-efficiency and near-zero infrastructure management. We leverage on exciting technologies like WebRTC, AWS Lambdas & Container Orchestration Service, and distributed data stores like MongoDB, SQS, Elastic Search, Redis etc.
The Video KYC solution has been augmented by industry leading straight through onboarding Karza solution- OCR and real-time KYC authentication APIs, passive activity detection and automated matching of names, addresses and faces. Powered by deep learning models and information retrieval algorithms, our models have been trained and tuned on millions of Indian names and addresses to match human-level performance in matching names and addresses across submitted documents and the application form data.
The in-house face matching Karza solution-which has been customized to handle quirks pertaining to Indian faces like turbans, etc – is used to compare a live photo, taken during the Video KYC process, with the photograph on an official document (PAN card and Aadhaar). To counter presentation attacks -where a fraudster may provide a non-live image or video to defeat the system, we also detect liveness of the user being authenticated. We provide robust video and passive Liveness solutions that answer the question – “Is this a real person?” and provide assurance about the presence of a genuine person.
Our VideoKYC is also equipped with OCR modules that allow data in official documents to be read and stored in a digitized format. Our state-of-the-art OCR solution enjoys: modified loss functions for localizing aligned documents, a training pipeline optimized for the BFSI sector by profiling frequently used documents for various font sizes, families, text densities and noises, Graph Convolution Network based parsing models to build an end-to-end OCR training pipeline with less than 50 training samples for any new document.
What is your market size in India? What has been the growth like in 2020?
Video KYC is potentially a $50Mn market. As awareness and adoption increases, we are seeing demand growing every month. Customer traction has increased by a 3-5x volume m-o-m.
This month IRDAI permitted insurance companies to use Video KYC. Additionally, with recovery factors like the removal of RBI repayment moratorium and clients build additional calling capacity, we believe these numbers should grow exponentially in the near future.