My work at Signzy

Joined in April 2019 as an intern, and it was my first taste of a real product-based software engineering company.

RM-Assisted AI-Enabled Onboarding Application for IndusInd Bank

My first days at signzy started as a 3-month internship. I was assigned to work on a project for IndusInd Bank, where I was maintaining and building new features for their RM (Relationship Manager) assisted AI-enabled onboarding application.

Soon, we worked on adding new features and integrations, more different types of entities, document scanning and verification and more.

This led to me being offered a full-time role at Signzy. Some of the technical breakthroughs I made during this project were:

• Stream Processing: There used to be data coming from multiple sources for various documents, and processing them one at a time proved to be inefficient and using basic NodeJS Stream handling largely yielded over 2x performance improvement.
• Backend Optimization: The backend used to face quite some traffic and I had optimized the backend in several places and used more efficient algorithms to improve the performance
• PDF Generation without a headless browser: The PDF generation used to be a bottleneck in the application, and I had replaced the headless browser with a more efficient library to generate PDFs. This led to over 10x reduction in memory usage while speeding up the PDF Generation process.
• CRM Integration: We integrated with CRMNext, a leading CRM provider where the onboarding data from our app went to the CRM application.

Internal API Team

Signzy had a lot of APIs for various purposes such as Digilocker and AADHAAR integration, and I was assigned to work on the API team. I was responsible for maintaining and building new features for the APIs.

Some of the technical breakthroughs I made during this project were:

• Digilocker Integration: I had built the entire Digilocker integration from scratch, this led banks and NBFCs to integrate these features a lot faster and more efficiently.

• AADHAAR Integration: From Aadhaar E-Sign to E-KYC, I had worked with stakeholders and industry leaders to build significant parts of the AADHAAR integration. Through this, customers can easily identify and onboard users in a frictionless manner.

• Rewriting the API Stack: We used to write APIs using express, which had the advantage of being simple, and the disadvantage of being memory intensive. We (me and 2 other colleagues) had rewritten a total of 200+ APIs, which resulted in over 25% savings in infrastructure costs.

• Face Liveliness Detection: Liveliness detection works on the user following some instructions that are displayed in-app such as:

  • Blink
  • Smile
  • Open Mouth
  • Nod
  • Shake Head
  • Turn Head (Left/Right/Up/Down)
  • Read alout a given text

  I had built the entire face liveliness detection system from scratch, which was used to detect if the user was following the instructions or not. We used Picoface and 68-point landmark detection and then writing rule-based algorithms to detect the liveliness. As we had to run these in very constrained environments at 10+ FPS, we had to optimize quite extensively, it was in 2020 and webRTC APIs were much more immature then.

• Fraud Detection and AML/CFT measures: I contributed significantly

Video KYC

Video KYC involves solving several key problems, some of which is regulatory and the others are technical and business oriented.

• Regulatory: Video KYC is a regulated space, and we had to comply with the regulations of the RBI and the KYC norms of the banks.
• Technical: Video KYC involves a lot of technical challenges such as:
  • Bidirectional Pairing: There are KYC agents and customers, and we had to pair them in a bidirectional manner such that the agent can see the customer and the customer can see the agent in a 2-way video call.
  • Remote Control: The agent should be able to control the customer’s device remotely to capture the documents and videos.
  • Media Streaming: We had to stream the video, and simultaneously record it for review. The video had to be in sync 100% of the time.

Video KYC solution for Citibank

One of our premier clients was Citibank, and we had to build a video KYC solution for them. I was the Technical Point of Contact and led several technical decisions and integrations.

• Compliance with Citibank’s regulations and norms.
• Complying to the InfoSec standards set in place.
• Integrating with Citibank’s existing systems.
• Building a scalable and robust solution that can handle the load of Citibank’s customers.