Vivek Paharia Email & Phone Number

San Francisco, California, Us

We are team responsible for creating inspiration for New Verticals consumers by showing them appealing items that reflect seasonal, local, and trending moments while providing a highly personalized grocery experience to each consumers' tastes and preferences. Our portfolio spans discovery, personalization and feed generation for New Verticals (grocery, convenience, alcohol, and retail storefronts).

San Francisco, California, Us

· Leading the Unauthorized Charge Detection and Prevention Service to save millions of dollars every year for the company· Developed ML models to detect the riskiness of the transaction along with the rule-based framework. Implemented various challenges to prevent bad actors and allow easy access to good users· Established data pipelines for historical and real-time features extraction, training and experimentation for ML models allowing rapid development and release of new models

Walldorf, Bw, De

· Established team of 14 members to take over the development of SAP Cloud Platform SDK for iOS which included Assistant Mac App, frameworks for mobile services and native UI for SAP Fiori · Worked on upgrading SDK to the latest iOS technologies by adding support for Dark Mode, CoreML and Mac Catalyst

Walldorf, Bw, De

· Leading the team responsible for machine learning offerings of SAP SDK for iOS. Provided the framework to run the models on the device along with the full lifecycle support from feedback submission to retraining using python-tf stack to redeployment using CoreML· Using tf-servings framework to deploy models at scale and fine-tuning algorithms to support customization of vision related tasks.· Developed and provided out of the box models and templates for various vision related use cases. Provided framework to help app-developer adopt the templates according to use cases.

Walldorf, Bw, De

• Architected SAP Content 2 Go system end-to-end by understanding business needs, transforming them to software development needs according to modern design trends and communicating them to backend, frontend and mobile team• Leading the development of the iOS and Android version of SAP Content 2 Go in team of 4• Strengthened SAP Mobile portfolio by designing and developing the SAP Content 2 Go iOS app single-handedly from scratch• Decreased inter-team dependencies by designing, developing, performance testing, building, releasing the app and undertaking customer engagement efforts ourselves• Developed MVVM style app heavily using Collection view and custom layouts with Core Data, GCD, URLSession and Webview• Added features such as Rich Push Notification, Spotlight search, keychain sharing, in-app search using Boolean retrieval, progress bars, location services, multiple languages, background sync and enterprise security for better user experience and intuitive user interfaces• Developed iOS iMessage App Extension, Today Extension and Quick Look Preview Extension along with support of 3D touch• Performed performance analysis to improve memory allocations, leaks and animation by adding image caching, layout caching and reusable cells & objects

Walldorf, Bw, De

• Working on various Java technologies like OSGi-runtime, Spring framework, Spring-MVC, RESTful web- services, JPA, JNDI and maven while prototyping new features for SAP Mobile Platform(SMP)• Evaluated SIP-SIMPLE, XMPP, MQTT protocols for encrypted-IM(Instant messaging) feature in SMP. Created end-to-end proof-of-concept(POC) of XMPP based IM service using Prosody as chat server, SMP as mobile platfrom and asmack as android client library.• Worked on Microservice architecture as part of next generation SMP features where services like IM would be running as seperate application and SMP would act as edge service.• Researched open-source solution to create Infrastructural Template for Microservice style application for HCP deployemnts consisting of a service discovery and registration service, a configuration service, client side load balancing and routing, inter-service communication library, circuit-breaker and resilience module, security & authenticaton mechanism and monitoring & logging module

• Created a framework for learning compact and fast image captioning models that can be deployed on smart phones and embedded devices which are limited on memory. Used knowledge distillation technique to train our small efficient model(student) from large and slow model(teacher). Reduced the memory footprint of model by 80% and caption generation time by 10% without compromising on metrics like perplexity and BLEU • Demonstrated Transfer Learning via Meta-Learning in Atari environment. We investigate the efficacy of Model Agnostic Meta Learning (MAML) on Pong and Breakout. We showed that within 3 gradient steps Agent which was trained on intrinsic task of just returning the ball was able to play the full games of Pong and Breakout.• Implemented semantic segmentation with transposed convolutions to identify the boundaries of a car in an image• Implemented attention-based sequence to sequence multi-layer RNN model to train chat bot on Twitter chats using v1.2 Tensorflow API. Inferences are made using beam search• Implemented character level language model using multi-layer RNN to generate Shakespeare like plays• Digit Recognition on MNIST using CNN. Tricks like Batch Norm, Xavier initialization, dropout, decaying learning rate helped to score 99.257\% on Kaggle• Predicted seizures based on EEG data using 2-layer NN purely implemented in numpy Applied extensive feature engineering to predict House Prices. Achieved loss of 13.17\% after comparing various models• Predicted major shopping streets/complexes on Yelp dataset using PySpark to help tourists find hot spots in various cities• Achieved 78\% accuracy in predicting Titanic survivors on Kaggle

Santa Clara, California, Us

• Developed infrastructure to enable automatic and easily configurable testing of clock synchronization and stream reservation protocols of Audio-Video Bridging stack on real and virtual network switches• Used Linux Network & PID namespaces for virtualization, configured Linux kernel to enable hardware time stamping and augmented benchmark to make it configurable using Linux named pipe

Gurgaon, Haryana, In

web backend development

Bachelor Of Technology (B.Tech.), Computer Science And Technology

Graduate Certificate, Artificial Intelligence