M. Zeeshan Siddiqui Email & Phone Number

Redmond, Washington, Us

AI Compiler team for Azure MAIA 100 AI accelerator.Leverage MLIR and LLVM to compile deep neural network graphs to machine code, while also working on hardware-software co-design.

Redmond, Washington, Us

AI Frameworks, Artificial Intelligence Platform.• Accelerated deep neural networks(DNN) training in ONNX Runtime (https://github.com/microsoft/onnxruntime) by improving ONNX spec, kernel implementation, static graph memory management, designed and implemented partial graph execution to enable ONNX Runtime to be used as PyTorch backend via ORTModule, authored "10,000 foot view of ORTModule internals" section at https://techcommunity.microsoft.com/t5/azure-ai/accelerate-pytorch-transformer-model-training-with-onnx-runtime/ba-p/2540471• Technical lead for ML .NET C# based machine learning framework (https://github.com/dotnet/machinelearning) from 05/2019 - 01/2020, mentored several new hires, managed interns, release manager for monthly releases, scrum master, led the design and implementation of several features from inception to shipping, collaborated with multiple internal teams and drove first party adoption within the company. Prior to this worked on improving accuracy and speed of ML algorithms(win rate), engineering new learners and transforms and ONNX runtime. Co-inventor on multiple patents and co-author on research publication.

Redmond, Washington, Us

Compute Platform, Azure Compute.• Improved Virtual Machine Scale Set (scalable group of VMs) reliability and availability guarantees, i.e creation success rate from ~50% to ~99%, spanning VMs across availability zones for higher availability. Co-inventor on multiple patents.

Redmond, Washington, Us

Kernel Platform Group, Windows Core Operating System Division (now known as Base)• Windows Subsystem for Linux 1 (based on linux system call emulation) that enabled bash to run natively on Windows.• Improved system performance by prefetching memory pages, wrote tool to debug heap, and contributed to microsoft edge document object model.

Austin, Tx

CS 341: Automata TheoryCS 303E: Elements of Computers and ProgrammingGraded homework and programming projects.Led discussion session once a week.

Sunnyvale, Ca, Us

Worked in Behavioral Targeting (Short Term Real Time Scoring), Display Advertisement and Marketplace Engineering group.Designed and Implemented a framework to generate generic data that was initially intended to create valid but dummy cookie records in the user data store but due to flexibility that the tool offered, it is now being considered for use in other areas at Yahoo! as well.-Tool works by having fundamental blocks that emit data which could be used in an hierarchical order to construct complicated textual structures.-At the same time, the tool offers room to add more custom designed blocks based on future needs and hence offering plenty of flexibility.-Ability to push data into the output file from an external source with a generic structure.-Configuration file driven, that is, user can specify the output structure using blocks in a JSON file and tool generates the data based on it.Implemented core back-end service to help in opting users out from being targeted for sensitive categories. Received fulltime offer during mid-internship to come back in 2012.

Austin, Tx

Fall 2010 : CS 315H Data Structures and Algorithms (Honors)Spring 2011: CS315 Data Structures and Algorithms-Conducted office hours thrice a week-Held discussion session twice a week.

West Lafayette, In, Us

Taught labs for:-Programming 1 (CS 180) from Spring 2008 - Fall 2009-Programming with multimedia objects (CS 177) in the Fall 2007. -Designed 2 labs every semester, proctored approximately 14 labs each semester or a total of 70 during my undergraduate days. -Taught over 100 students graded labs for each student and maintained a grade record.

Seattle, Wa, Us

Worked in Website Application Platform.Designed and implemented a reporting system on top of a dependency graph framework that allowed the following.-Generation of generic reports at fast pace-Scheduling, pre-rendering and cacheing time intensive reports to save time.-Easy way to communicate between several data service by providing plug-able service adapters-Users could fire up or alter a report simply by writing an XML file that would instantiate java beans instead of writing any actual code.

Santa Clara, California, Us

Worked in Advanced Visual Computing Group.-Designed, implemented and optimized a library call that would synchronize clock times between any two cores(CPU and GPU) at a given instance in time to a precision of up to 99.97%.-Wrote a few benchmarking tools to report the throughput and bandwidth of the library calls used to communicate between two processes.

Bangalore, Karnataka, In

Worked in Software Engineering and Technology Labs (R&D)-Wrote a tool that would compare two graphs and report which nodes were changed and indicate appropriate branch that needs to be tested (The graphs represented control flow diagrams of two versions of the same software)

Ms, Computer Science

Bs, Computer Science Honors