Krish Patel - Software Engineer II | MSc CS Researcher, Waterloo

AboutMe

I'm Krish Patel, a Software Engineer II at Huawei's Distributed Data Storage & Management (DDSM) Lab in Markham, and an MSc (thesis) student in Computer Science at the University of Waterloo, researching under Professor David Toman in the Data Systems Group at the Cheriton School.

Before Waterloo I completed an Honours B.Sc. in Computer Science (High Distinction) at the University of Toronto, where I also spent three years as a teaching assistant across nine CS and math courses and worked as a research assistant on speech-interactive learning tools. If your work sits where deep systems engineering meets ML, I'd like to hear about it.

MyEducation

MyExperiences

ProfessionalExperience

Reduced IVF vector index build time by 35% by implementing asynchronous column unit (CU) prefetching that warms the CU cache with the units a scan is about to read, hiding disk latency on I/O-bound index build and query scans.
Improved vector index create time by 70% and throughput by 40% queries per second on a standard vector database benchmark by adding Parquet page index (ColumnIndex and OffsetIndex) support to the storage reader.
Enabled direct page-level reads on data sets up to 10 TB, where a column unit cannot fit in the memory cache, by seeking to individual Parquet data pages via their byte offsets and skipping pages whose min/max statistics cannot satisfy the predicate, instead of decoding whole column chunks.

Accelerated database OLAP query engine by 53.09% by developing a C++ constant folding optimization that replaces scalar subqueries with literal values in the query rewrite stage, validated against the TPC-DS (SF100 benchmark).
Built vector support in Huawei's proprietary XStore engine, enabling storage and retrieval of 2 embedding types for vector search workloads.
Added BM25 full-text indexing to the XStore engine, enabling relevance-ranked text retrieval and expanding native search functionality by 1 index capability.
Extended the datalake/formation and DiskLPAS indexing infrastructure to support 4 index types across create, rebuild, load, search, insert, and delete workflows.

Set up Jenkins pipelines for the team, automating compilation and recovery runs, resulting in a 90% reduction in manual setup time, from 20 minutes to 2 minutes, allowing team members to focus on other tasks.
Improved database performance by 36.67% by implementing sub-query caching with materialized views, measured across the TPC-DS and TPC-H benchmarks.
Improved GaussDB's recovery speed by 61.53% by optimizing system calls and reducing read operations, increasing recovery speed from 130 MB/s to 210 MB/s on average across 10 runs.
Enhanced SPSCBlockingQueue performance by 42.8% through caching optimizations, achieving 10M operations per second, outperforming Facebook's folly queue, as verified through standalone benchmarking scripts.
Optimized thread affinity by balancing distribution across NUMA nodes, which led to a 50% increase in WAL recovery speed, improving from 400 MB/s to 600 MB/s in a distributed database environment.

Built a data-scrubbing feature using NLP models to redact Personally Identifiable Information (PII) and Protected Health Information (PHI) from text, PDFs, and images, increasing data privacy compliance by 70%.
Developed a native messaging Chrome extension to scrape and transfer webpage HTML for task automation, improving LLM task prediction accuracy by 25%.
Automated installation for OpenAdapt with a comprehensive PowerShell script that sets up dependencies across platforms, reducing setup time by 40% and increasing deployment consistency.

TeachingExperience

Graduate teaching assistant for CS 115 (Introduction to Computer Science 1) at the David R. Cheriton School of Computer Science, the introductory course on program design and the fundamentals of computation through functional evaluation, with programs written in subsets of Scheme.
Lead tutorial and lab sessions, guiding students with no prior programming experience through recursion, compound data, and recursive data definitions.
Hold weekly office hours and answer student questions on the course Piazza forum, covering both course concepts and debugging help.
Mark assignments and exams, providing structured feedback and consistent grading across the section.

Completed 18 TA appointments across 9 unique Computer Science and Mathematics courses: CSC108, CSC148, CSC207, CSC209, CSC263, CSC343, MAT135, MAT136, and MAT232.
Supported 2000+ students across programming, software design, systems programming, algorithms, databases, calculus, and linear algebra.
Led TA teams of 8 to 10 members for CSC207, CSC343, and MAT232 across 5 terms, coordinating weekly tutorials, grading workflows, and consistent student support for 200+ students per course.
Delivered tutorials, held office hours, graded assignments and exams, and created educational content including walkthrough videos for OS memory management and guided problem solving for calculus topics.
Ran course operations using CrowdMark, Piazza, Canvas, and MarkUs.

Taught coding and robotics to students from Grade 1-12, tailoring lessons to different skill levels while fostering creativity, problem-solving, and team collaboration.
Designed engaging STEM projects (Python games, Arduino robotics) that improved students' coding proficiency and sparked interest in technology.
Developed individualized support strategies that improved student participation and retention in the program.

ResearchExperience

OtherExperience

MyProjects

Snowfall Saga (Multiplayer Web Game)

Built a real-time multiplayer web game with WebSocket-driven state synchronization between players, and streamlined releases with GitHub Actions and Docker.

Snowfall Saga (Multiplayer Web Game)

Built a real-time multiplayer web game with WebSocket-driven state synchronization between players, and streamlined releases with GitHub Actions and Docker.

DataGuard (PII/PHI Data Anonymization Tool)

Built Scrub, a Python tool that detects and redacts personally identifiable and protected health information across text and images, backed by NLP models and a regression test suite for detection accuracy.

DataGuard (PII/PHI Data Anonymization Tool)

Built Scrub, a Python tool that detects and redacts personally identifiable and protected health information across text and images, backed by NLP models and a regression test suite for detection accuracy.

Portfolio (This Website)

Designed and built this responsive, SEO-optimized personal site in Vue 3 and TypeScript, continuously deployed to Netlify, with structured data and generated icon and social-card assets.

Portfolio (This Website)

Designed and built this responsive, SEO-optimized personal site in Vue 3 and TypeScript, continuously deployed to Netlify, with structured data and generated icon and social-card assets.

Chronos (Time Management System)

Built a web-based time management system in Java, applying SOLID design principles, UML modeling, and Scrum ceremonies across the delivery cycle.

Chronos (Time Management System)

Built a web-based time management system in Java, applying SOLID design principles, UML modeling, and Scrum ceremonies across the delivery cycle.

MySkills

Programming Languages

Python

C++

JavaScript

TypeScript

Java

C

PHP

Assembly

Haskell

Racket

JavaScript Frameworks/Libraries

Express.js

React

Node.js

REST API

Vue.js

Ajax

Next.js

jQuery

Databases

MongoDB

GaussDB

PostgreSQL

MySQL

openGauss

Redis

SQLite

Neo4j

AI/ML

Deep Learning

Numpy

Pandas

OpenAI

PyTorch

TensorFlow

Hugging Face

Scikit-learn

SpaCy

Keras

OpenCV

Scripting Languages

Bash

Shell

PS1

Groovy

Version Control

Git

GitHub

GitLab

Code/Text Editors

VS Code

Vim

Notepad

DevOps Tools

Docker

Slack

CI/CD

Jenkins

Azure

Selenium

Jira

Kubernetes

Terraform

Markup/Modeling Languages

HTML

Markdown

CSS

TailWind

XML

LaTeX

SSML

Operating Systems

Linux

Windows

Android

Unix

macOS

CentOS