Jay L Karp

jay karp

Hi, I'm

Jay.

Cartoon drawing of Jay at his desk, drawn by Jay

Word cloud about current interest, topics include: languages, systems, frameworks, etc.

I’m a software engineer working at Twitch as part of Amazon Web Services IVS. My interests span the full product development process, from user research and prototyping to system implementation and design. When I’m not programming, you can find me playing saxophone and piano or baking fresh bread. I’m currently looking for job opportunities for 2023. Check out some of my projects below - including the website itself! And please contact me with any questions or opportunities.

Languages.

Golang

Typescript & Javascript

HTML & CSS

Swift

Haskell

Ruby

Bash

LaTeX

Frameworks.

React

Styled UI

Node.js

GraphQL

Bootstrap

JQuery

Apollo

SwiftUI

AppKit & UIKit

NumPy

Selenium

Flask

Tensorflow

Ruby On Rails

Parse

Technologies.

Vim

TMUX

VS Code

Microsoft Office

Git & Github

G Suite

Docker

AWS

Firebase

Balsamiq

Figma

MailChimp

Experience

twitch.

The backbone that powers twitch's video infrastructure is an AWS team called Amazon IVS (or interactive video services). As an engineer within Amazon IVS, I work on the interactivity team developing and architecting high-quality features that provide interactivity to a customers video stream. I am responsible for building new features using AWS infrastructure best practice to provide customers with operational excellence, and responsible for optimizing the performance, reliability, and scalability of our cloud systems.

GOLANG.TYPESCRIPT.AWS.CDK.PYTHON.OPS

Cloud Engineering and Architecture

Operations

apple.

During my time at Apple I built tooling for the Customer Feedback Engineering team. I created custom designs for different user interfaces by following an iterative design process which consisted of talking to users and creating both paper prototypes and wireframes. After completing the design process, I developed the designs into a fullstack application using a Ruby on Rails backend with the GraphQL Ruby gem. I then used an Apollo middleware to fetch data from the backend. Finally, I built the frontend application based on my previous designs concepts using SwiftUI as well as specialized AppKit views.

SWIFT.RAILS.SWIFTUI.APPKIT.UIKIT.GRAPHQL

UI & UX

Fullstack Dev

ceal.

CEAL (Columbia Computer Enabled Abilities Lab) is an interdisciplinary lab enabling those with differing abilities to experience the world. Our research spans accessibility, games, AR, AI, sensing and vision. Our website serves dual purposes: recruiting researchers to grow the lab and enrolling those with differing abilities in our studies. We want to ensure our lab’s interface is reflective of the participants we work with, as typical media connections sometimes rely on inaccessible audio or visual communication. My contributions to this website spanned the entire development process, ranging from early prototyping and design to implementation. We expect to be live at the end of January 2021.

HTML.CSS.JAVASCRIPT.PYTHON

UI & UX

Fullstack Dev

navstick.

Navstick is a CEAL lab project exploring accessibility for people with visual impairments. Using self-directed audio navigation, we aim to encourage autonomous shopping. Existing supermarket navigation solutions are cumbersome, and ours allows users to explore the shelves through self-directed verbalization of the environment. We created a supermarket simulator game for our virtual testing environment and a remote user research framework. During the course of the project, I ran 15 user studies over Zoom, built web interfaces to control the game environment remotely, and built the database/backend design to enable communication during testing. Upon the completion of our prototype, testing, and evaluation, we submitted two papers to UIST and CHI. The lab is further refining phases based on our findings.

REACT.PYTHON.BASH.C#.UNITY.PARSE

UI & UX

Fullstack Dev

cheddar.

Cheddar is a finance and tech based streaming news network with it’s trademark 24 hour live news coverage. As a constantly live site, Cheddar has unique opportunities and challenges, especially in tool development and deployment. I worked on several projects over the course of my internship, gaining experience in many different areas of online news delivery production. The key area of my internship was developing tools to ensure a seamless 24 hour news broadcast. Two of the projects that I focused on were monitoring web content and content distributed over a raspberry pi distribution network. My contributions varied from writing code for a golang api in the prometheus pod for kubernetes cluster, to writing software to update the raspberry pi network remotely. I also enabled highly requested features on the site, including internal tooling to connect articles and highlight authors, as well as developing the tooling required to launch their Need2Know newsletter.

REACT.GOLANG.RUBY

UI & UX

Fullstack Dev

DevOps

vanderbilt.

I explored three main projects during my time at Vanderbilt in data science and biomedical research. First, a joint venture in electronic health records with EPIC enabling verbal interactions with patient information documentation. This increases efficiency by allowing doctors to notate while occupied and reduces search time for patient information. To understand doctor’s perception of NLP tools, I built test tools to enable large-scale user testing via nlp querying tools to compare EPIC to traditional voice assistant models, trained on medical and not medical words. Second, our lab investigated reversing diabetes in mice using cancer therapy drugs. We examined different transcription factors and their ability to reverse diabetes by performing immunohistochemistry, Western blotting, dissection and husbandry techniques. Lastly, I calculated the distance of various zip codes and their closest hospital for UNOS organ donation allocation policies. Both the transplant allocation work and transcription factors research resulted in published papers, viewable in my resume.

JAVASCRIPT.JAVA.PYTHON

NLP

Statistics

UI & UX

Projects

musely.

store.subscribe(() => console.log(store.getState()))
ReactDOM.render(
  <Provider store={store}>
    <Router>
      <App/>
    </Router>
  </Provider>,
  document.getElementById('root'))

a music note taking tool for the average student

UI & UX

→

par-sort.

shuffle :: V.Vector a -> IO (V.Vector a)
shuffle v = do
 let n = V.length v - 1
 js <- V.forM (V.enumFromTo 0 n) $ \i -> randomRIO (i, n)
 return $ V.create $ do
    o <- V.thaw v
    forM_ [1..n] $ \i -> M.swap o i (js!i)
    return o

parallel functional sorting implementations in haskell

Parallel

Functional

→

paxos checker.

p1PreparePhase := func(s *base.State) bool {
  s1 := s.Nodes()["s1"].(*Server)
  valid := s1.proposer.Phase == Propose 
  if valid {
    fmt.Println("... p1 entered Propose phase")
  }
  return valid
}

distributed, stateful paxos error checking

Distributed Systems

→

Musely.

In fall of 2019, I worked on Musely: a musical note-taking app. Our initial user studies exploring pain points for music students revealed a difficulty recording information about music - especially when those notes were tied to a certain time in the piece. Previously, a student would need to write down a timeframe and musical description (eg: adagio) and later match it to the piece. This written description was very difficult for students to relate to a piece of music, especially in remembering how the verbal notes related to the music without re-listening. Our solution integrates the music and the annotations in the same interface, no longer a separate audio file and paper time markings. We built a prototype that enabled students to notate as they listened to the music, and to see a visual representation of their notes overlaid on the audio file.

REACT.JAVASCRIPT.REDUX.BALSAMIQ.FIGMA

par-sort.

Through this project, we aimed to explore the possibilities of parallel sorting in a functional language. With the possibile slowing of Moore’s law in the near future, it is essential that we look at alternative methods to push these important algorithms even further. In our research, we implemented three main sorting algorithms using Haskell: merge sort, quicksort, andbitonic sort. Although it would probably be faster to write these sorts in a lower level language like C, we believe that the functional paradigm allows for much easier and more concise data parallelism. After moving these three main sorts from their equivalent sequential and imperative counterparts, we further pushed the limits of sorting in Haskell to implement some of the different data parallelism models. Finally, we present ‘hybrid’ sort, which is composed of the best features from all of the sequential, and parallel sorts that we implemented.

HASKELL.STACK.THREADSCOPE

paxos checker.

When we build a distributed system, we check if it works properly in all cases, even some very low probability corner cases. For this project we aim to prove the correctness of an implementation of paxos. We evaluated two different methods: model checking and fuzz testing. Fuzz testing occurs when you repeatedly test multiple different randomly occurring states, and is efficient in practice. However, there is no guarantee that this randomized testing can cover every case. Instead we pursued the Model checking approach, exploring every possibility, potentially achieved by simulating a system by a state machine. Then search algorithms like Breadth First Search (BFS) or Depth First Search (DFS) can be used to walk through all the states and verify the correctness of one's implementation. In this assignment, we considered the whole system (nodes + network) as one large state machine. Each node represented a client or server that could send or receive messages. The network represented an abstraction of on-the-fly messages being transferred over a real network. We first implemented the state machine as well as the different state transitions. We finally wrote our model checking tests, and verified that our solution was viable, successfully modeling paxos.

GOLANG.PYTHON.BASH

Connect

designed and built by me.