David Wu

Senior Software Engineer

Backend Systems • Automation • Applied AI

David Wu is a New York based senior software engineer focused on backend systems, applied AI, internal platforms, and developer tools.

New York, NY

Experience

Senior Software Developer

Taboola · New York, NY

Feb 2023 — Apr 2026

  • Led applied-AI work for account review, shipping a categorization system that reduced manual review volume by 40% with human-in-the-loop validation
  • Designed and piloted an MCP-based LLM assistant for ~20 internal users, unifying content generation and real-time analytics workflows
  • Helped move legacy monoliths into Kubernetes microservices while owning 5+ content-review applications used daily by ~100 reviewers

Internal Applications Technical Lead

Taboola · New York, NY

Feb 2020 — Jan 2023

  • Re-architected the core internal platform from a single server into a horizontally scalable distributed system, raising uptime from 98% to 99.8%
  • Led a Grails 2 to 3 migration while increasing automated test coverage from ~20% to ~80%, improving release confidence
  • Managed and mentored two engineers while introducing sprint planning, standups, and structured code review

Internal Applications Developer

Taboola · New York, NY

Apr 2017 — Feb 2020

  • Built self-service tooling for account management and support teams, reducing technical-support escalations by ~70%
  • Integrated Salesforce and JIRA APIs into campaign workflows, reducing approval timelines from ~2 days to same-day turnaround
  • Created Chrome extensions for in-browser ad placement editing, removing manual design handoffs from routine operations

Skills

Languages

Python, JavaScript, Groovy, Java

Frameworks

Grails, FastAPI

Platform

Kubernetes, Docker, Helm, Jenkins, GCP

Data & Messaging

MySQL, Vertica, Redis, RabbitMQ

Observability

Prometheus, Grafana

AI & Integrations

OpenAI API, MCP, Salesforce API, JIRA API

Testing

Spock, JUnit, Geb, Selenium

Web

HTML, CSS

Libraries

React, jQuery

Projects

telemouse

Record raw mouse input during play, replay the session, and inspect movement and aim metrics in a browser.

Built the Rust capture agent, browser visualization, session recording, and offline analysis tools.

Batch input reads to reduce CPU work, with an exact-timestamp mode when timing precision matters more.

Capture-thread CPU fell from 2.8% to 0.22% of one core at 1 kHz on a Ryzen 5950X with an 8 ms coalescing window.

Capture benchmark and tradeoffs

Rust, Windows Raw Input, UDP, WebSocket, Kafka

vidgrep

Find text or visual patterns in recorded video, then export matching timestamps or clips. A resumable worker handles larger libraries.

Built the scan and clip pipeline, resumable batch worker, and tools for inspecting OCR results.

Profile decode and OCR separately; use hardware decoding when supported and fall back to software on other machines.

A published RTX 5090 benchmark reduced a 26.7-second AV1 video scan from 29.6 s to 5.1 s, preserving all eight search matches. Results depend on codec and hardware.

Benchmark setup and verification

Python, FFmpeg, OpenCV, OCR

LocalScribe

Transcribe audio and video locally on a GPU, with word-level timestamps and optional two-speaker labels. Runs as a CLI or as an MCP server that watches a folder.

Python, WhisperX, PyTorch, MCP

CandleTunes

Turn stock and crypto price history into multi-track music, synced to candlestick charts and an on-screen piano. Mix tracks, pick a scale, and export MIDI or audio.

FastAPI, React, TypeScript, Tone.js

silksongRL

Train a PPO agent to beat a Hollow Knight: Silksong boss from raw pixels. A BepInEx mod streams exact HP over UDP, so the reward is precise and unfarmable.

Python, Stable-Baselines3, PyTorch, Gymnasium

Education

CUNY Brooklyn College

Bachelor of Science, Computer Science

New York, NY · 2011 — 2016

Contact

hello@davidwu.com

GitHubLinkedInX / TwitterResume