Sunny Singh

Working on a multi-tenant GKYC SaaS platform that provides KYC/KYB workflows as a service for large wallet and financial clients. The platform is built as a distributed, event-driven system on AWS and currently powers OVO Business (Grab-acquired identity platform), with ongoing expansion to Grab PH. It handles end-to-end KYC flows including initiation, submission, risk evaluation, auditing, and manual review, operating at ~100k daily KYC initiations in production.

At Grab Financial Group(GFG), have been a part of the 𝐂𝐨𝐦𝐩𝐥𝐢𝐚𝐧𝐜𝐞 Tech Family, contributing to Merchant KYB and Passenger/Driver KYC flows, primary objective being to compute & provide KYB/KYC Level used before each GrabPay transaction. GKYC Portal (Grab Defence): Currently working on building a SaaS: KYC Portal, for all kyc/kyb needs for clients. 𝐌𝐞𝐱 𝐎𝐧𝐠𝐨𝐢𝐧𝐠 𝐃𝐮𝐞 𝐃𝐢𝐥𝐢𝐠𝐞𝐧𝐜𝐞: • Primary contributor in designing the Merchant Re-KYB flows for re-evaluating the risk of all the ~400k Merchant profiles registered across Singapore and Malaysia on Grab-Pay/Food/Mart. Collaborated with Cross Functional Teams, lead discussions and regularly presented progress reports to stakeholders, delivering the product in time. • The project involved Exposing REST endpoints, writing State Machines, Lead Streamers, Schedulers, Migrations, Salesforce Integration, Several Vendor Integrations, Secrets Handling, Caching, UTs, Building Mex ODD UI in React to support Risk-Ops and Merchant-Ops Teams, etc. 𝐌𝐞𝐱 𝐄𝐧𝐡𝐚𝐧𝐜𝐞𝐝 𝐑𝐢𝐬𝐤 𝐑𝐚𝐭𝐢𝐧𝐠: • Closely worked with product to align on the requirements, formulated HLDs, LLDs to build a Risk Evaluation Framework to evaluate all Grab Merchants at various parameters set up by compliance to ensure that Grab Adheres to the GRCC & Tech Governance Obligations. The project involved Exposing REST endpoints, writing State Machines, Migrations, Backfill Workers, Kafka Producers and Consumers, UTs, Building ERRF UI in React etc. 𝐓𝐞𝐜𝐡 𝐒𝐭𝐚𝐜𝐤: Golang, MySQL, Kafka, SQS, Docker, kubernetes, Consul, Redis, AWS, Datadog, React, AWS Step funcs and Resux

• Rebuilt the preowned car pricing model using adv polynomial-regression techniques that reduced the poly-transformed feature count by almost 77%, improved model MAPE by about 3% from the previous release and brought the desired car price depreciation characteristics against year-odo and year-owner input variables. • Built a new car-price-governance model that dominated the previous existing heuristics and improved the MAPE of the predicted prices by 11%. • Improved Latency in car-price-prediction endpoints by masking transformed model coeffs. over non-transformed models, which allowed skipping feature drop processing and also restructured the Flask API endpoints. • Devised and deployed the ETL + MLOps Pipeline in Airflow for OLA-cars that was responsible for Extracting Data from relevant sources, training the models(pricing and governance), testing it and then post validation deploying it and alongside mailing the successful iteration report with various metrics and visualizations to the DS Team.

• Designed and Developed an automated annotation software customized for Autonomous-Drive Domain Datasets that allowed annotations ranging from bounding boxes, instance segmentations to 3D-fragmentation of the objects. • Built Flask API endpoints that provided bounding boxes, polygons and object segmentation boundary coordinates on input images. The application Backend was built in Node with Express as a framework and MongoDB for storing the annotation item coordinates. • Built the client-side frontend using React and the connected all these microservices together to have the application running, though not as a stand alone.

• Wrote Spark Jobs for PrimeTime analysis and the design validation that ran on the Mesos Cluster running on machines borrowed from the Synopsys Compute Farm. • Migrated Synopsys timing, power and primetime timing analysis workflows from LA(Library Analyser) Shell to Airflow DAGs. In Static Timing Analysis, improved the time complexity using multiprocessing. Similarly, In Dynamic Timing Analysis, moved the entire processing onto a Apache-Mesos cluster and wrote custom Frameworks and Executors in Mesos, thereby improving the Time Complexity of analysis in multiples. • Refactored and Modified the Senna Library Analyser PyScript into a Python Package and then migrated the entire workflow to Airflow. Primary Client: Intel

WebTeam is responsible for developing, maintaining and updating the Institue's website. Currently, We are building a new website for our institution using React.js and Node.js. I function as a full-stack developer.

I co-annotated inorganic trash images and co-developed a YoloV3 implementation in TensorFlow to detect and classify waste objects present in the trash and assign them to pre-decided classes. The model returns the relative coordinates of objects on the conveyor belt and the count of observed detections per class, using DeepSort Algorithms and concepts involving Image processing, Data Processing. And finally, deployed the model on Google Cloud Platform. Also held and attended regular meetings according to AGILE (Scrum) methodology.

It was a great experience working under Pratik Dhaboo Sir, at Awishcar.com. I was responsible for the development of Snapcart POS application, here I developed an Angular 8 Frontend and REST API backend with NodeJs which is agnostic and can handle all types of requests, including import and export of CSV files.