Jaideep Kukkadapu

Building QAIOR, Qualcomm’s Agentic orchestration framework for on-device AI on Snapdragon mobile and automotive platforms.

Teaching Assistant for the course CMPSC 461 - Programming language concepts

CiteSeerX (https://citeseerx.ist.psu.edu/) Part of Core Team : https://csxstatic.ist.psu.edu/people/team.html + Infrastructure, Full-stack + Design, Implementation, Monitoring, Testing + Collaboration, Ownership, Team work 1. Developed a sophisticated Machine Learning algorithm Learning to Rank (LTR), significantly improving the precision of search results within the CiteSeerX platform. 2. Implemented a series of optimizations resulting in a remarkable 10x improvement in search engine speed, reducing the response time from 6.2 seconds to 0.83 seconds. By strategically sharding and fine-tuning multiple indexes in Elasticsearch, we achieved super performance enhancements across the platform.

Part of the Qualcomm AI Runtime (QAIRT) team working on On-Device ML : 1. Improved the scheduling of subnets in neural networks to achieve true parallelism across heterogeneous compute systems (CPU, GPU, NPU), maximizing computational efficiency and throughput. 2. Reduced CPU involvement in buffer transitions between NPU and GPU, enhancing efficiency by leveraging memory-mapped buffers and eliminating tensor data copy costs during inference.

Developed and implemented a comprehensive ML SD-WAN dashboard using React, Prometheus, and Perl, automating data retention and management processes for improved network monitoring, analysis, and optimized system resources.

Worked as a research assistant in several deep learning projects that focused on segmentation and visual question answering (VQA).

Proposed, designed and prototyped a Software design to extend Qualcomm’s existing on-device inference framework - QNN to support on-device training. Implemented modifications to the QNN API, including back-propagation kernels and loss layers for CPU and GPU.

Used unreal engine to develop and simulate white board experiences to practice social engineering skills. Designed and developed a feature to render pdfs/ppts as slides in VR before a simulated audience.

• Coached 150+ students about mathematical concepts related to machine learning. • Conducted doubt-solving sessions and graded assignments.

Evaluated existing Deep learning solutions and implemented a robust solution for repetition count tracking with Posenet. Evaluated cloud vs edge inference, live vs offline repetition counts and created an online room like zoom where users can competitively perform group exercises and can see live scores of number of calories burnt.

Algorithms | Distributed systems | vision and language | Concurrent matrix computation(Parallel programming CUDA) | Advanced Operating Systems