David K. Email & Phone Number

Los Angeles, CA, US

Bethesda, Maryland, US

Los Angeles, CA, US

• Greatly reduced manual review of surface grading tasks by deploying new ensemble model with >90% generalized prediction accuracy for card surface condition inference on fastAPI endpoint - prior implementation suffered.
• Improved grading admin tool page load time by 200ms by migrating card image database calls from GCP bucket to imgix CDN.

• Contributed preprocessing module to TensorFlow.js. Layers include center-crop, category encoding, rescaling, and resizing layers.
• Contributed bicubic interpolation backend for TensorFlow.js. Standardized API to conform to Python-Keras implementation of TensorFlow, improving interoperability between the Python & JavaScript versions of TensorFlow..
• Directed implementation of BaseRandomLayer, and the derived RandomWidth and RandomHeight preprocessing layers.

• Predicted user-to-customer conversion likelihood based on demographic and platform usage data with a gradient-boosted classification model.
• Aggregated and centralized data from HubSpot CRM and platform usage metrics pipelined to Preset SQL database.
• Automated scheduled data fetch and centralized data using Cron with Selenium and Beautiful Soup-based web scraping data pipeline
• Created machine learning instructions, code documentation, and pytest-based unit tests for technical audiences within the organization.
• Improved internal workflows and materials, including presentation slides, for non-technical audiences within the organization.

Venice, California, US

• Applied Jest and Supertest to programmatically generate and run syntactically sensible unit and integration tests for imported codebases
• Used Electron.js to serve the Spearmint application in a single-window Chromium app independently from the local browser and OS while allowing the Node.js application to access system files and resources, enabling the.
• Utilized React’s Context API and hooks to modularize application state management and avoid global state pollution
• Migrated existing codebase to TypeScript for stronger typing to enforce stricter data contracts
• Built-in Google and Facebook OAuth as authentication alternatives, integrated bcrypt at login to hash passwords before storage
• Implemented and launched software distribution site hosted on AWS EC2.

Berlin, DE

• Fetched weather and air quality data through the weather API to create a no-code system for users to define rules and triggers for alerts based on weather and pollution values
• Utilized PostgreSQL backend to store rulesets and user information with bcrypt encryption in an ACID compliant and secure fashion
• Implemented session cookies stored on clients for durable login sessions, allowing persistent logged-in state across discontinuous visits

Berlin, DE

• Used React and Redux to create a stateful and responsive frontend for a Kanban-style task assignment board in which users can create or take on requests
• Utilized React-Router Hooks for seamless and programmatic redirects between pages on user interaction with rendered elements

• First data scientist at an ecommerce wholesale apparel company with annual revenue of ~$10M.
• Centralized and automated sales and production data scraping using Selenium, bypassing authentication with headless browser to automatically refresh data
• Developed regression model on one-hot encoded products & orders data to highlight SKUs where demand surpassed current inventory
• Displayed statistics and predictions using Microsoft Power BI for streamlined data pipelining and user-interactive charting
• Incorporated RESTful UPS shipping API to ERP system, automating shipping label generation & printing from invoiced items' weights in database
• Significantly reduced warehouse item retrieval time by deploying a custom script for calculating overlaps across different orders' items

• Integrated Huobi, Binance, BitForex, ZG.com RESTful APIs into Swift application to include broader market and wallet support
• Refactored Java Android version of application into Swift for iOS release
• Worked within a team in an Agile Scrum environment, taking advantage of planned sprints to meet delivery deadlines

San Diego, US

• Applied word2vec algorithm on a record of signaling events to determine when dynamics of the neural network have entered steady state as part of neuromorphic computing research
• Synthesized a literature review paper on organic sensory neural circuits of various animals and using that data to define a layer-wise sensory convergence ratio

• Experimented with multiple different data filtering and processing treatments for undersea pressure sensory data to train a deep learning network trained to determine ship locations from engine noise in environments.
• Presented the results of training the multilayer perceptron model on PCA-applied data, which could accurately predict ship location by sound in an environment with signal-to-noise ratio as low as -10.
• Research funded by grants awarded by the United States Navy.

• Utilized A/B testing for accessible UI/UX web design centered on main site and web store platform for wholesale clothing label to create a front-end which generated improved clickthrough and conversion rates
• AWS EC2 web deployment utilizing automated containerization and version control through Dockerized containers on ECR, SQL backend on RDS
• Filed company trademark with USPTO

Los Angeles, CA, US

• Built sensors and programmed data-taking & imaging steps in MATLAB for plasma generator machines used for studying acoustic/whistler waves.