Skip to content
Yue Wong

Full-stack AI engineer building systems that hold up in production.

Production MCP servers on live financial data, APIs that stay fast under load, and the infrastructure underneath.

MCP server shipped in under a week
7 tools
data-related defects
−85%
API latency
9s → <2s
request throughput
10×

Work

Personal projects

About

I'm a full-stack engineer at Heat Waves, where I build AI tooling for an enterprise finance platform. Outside work I run a small AI agent platform on a mini PC at home, and I build things for people around me when a simple tool would save them time.

I'm based in New York City. I'm a tinkerer at heart and like being hands-on, and I'm always looking for new problems to solve, especially the ones worth solving. Away from the keyboard, I'm a fan of all combat sports: I currently box and train Muay Thai, and I've done Brazilian jiu-jitsu before. I also love language learning, across Mandarin, English, Cantonese, and Spanish.

Experience

  1. −85%data-related defects
    • Beforebaseline
    • After−85%
    9s → <2sAPI latency, then −80% more
    • Before9,000 ms
    • Queries + indexes<2,000 ms
    • + Redis, async queue−80% more
    10×request throughput
    • Before1×
    • After10×
    • Led a production remote MCP server giving LLM agents secure, read-only access to platform data.
    • Built an LLM evaluation loop: a 25-scenario golden dataset, LLM-as-judge plus deterministic scoring, and production traces fed back into the suite.
    • Built a cross-database analytics microservice (Snowflake + PostgreSQL) with a custom discrepancy-detection algorithm.
    • Cut API latency with query consolidation and indexing, then a Redis cache and async task queue.
    • Feature-flag infrastructure end to end (Prisma, NestJS API, React context): progressive rollouts and kill switches without redeploys.
    • Kubernetes deployment for a monitoring service (probes, SOPS-encrypted secrets, Nx CI/CD) plus a BullMQ admin panel.

    NestJS · React · Prisma · PostgreSQL · Snowflake · Redis · BullMQ · Kubernetes · Nx

  2. +40%sign-up interaction
    • Beforebaseline
    • After+40%
    −30%deploy time
    • Beforebaseline
    • After−30%
    • Re-architected the sign-up flow for a more intuitive journey, lifting user interaction and satisfaction metrics by 40%.
    • Designed CI/CD pipelines on AWS CodePipeline, CodeBuild, and CodeDeploy, cutting deploy time by 30%.
    • Built and deployed new marketing websites: responsive layouts with MUI, plus AWS SES emails for profile validation and confirmation.
    • Served static content from S3 through CloudFront distributions, reducing latency and improving scalability.

    CodePipeline · CodeBuild · CodeDeploy · SES · S3 · CloudFront · MUI

  3. −50%deploy time
    • Beforebaseline
    • After−50%
    • Optimized Docker containerization with custom images and YAML configuration, halving deploy time.
    • Designed a custom Synthetic Canary against VPC-enclosed application endpoints, collecting real-time performance metrics into AWS CloudWatch dashboards.
    • Built a mobile-responsive interface in JavaScript and SASS with Flexbox and Grid.

    Docker · CloudWatch Synthetics · VPC · JavaScript · SASS

    • Migrated the codebase to TypeScript, catching type errors at compile time instead of runtime and making the code easier to maintain.
    • Designed custom Zustand middleware that snapshots the initial state and every change after it, then uses those snapshots to auto-generate frontend test suites.

    TypeScript · Zustand

Education

  • Georgia Tech

    M.S. Computer Science (OMSCS), expected June 2028

  • Hunter College

    Macaulay Honors College, Magna Cum Laude · New York, NY

  • AWS

    Certified Cloud Practitioner, 2023