San José, Costa Rica

Michael Chen Wang
Engineer & ML Researcher

Senior Software Engineer at Microsoft, building cloud systems at scale. Published at GECCO ’26 on evolutionary optimization for medical vision-language models.

9+

Years Engineering

1

Peer-Reviewed Paper

ACM

GECCO '26

Read the ResearchView ProjectsGet in Touch

About

Engineer & Researcher

I'm a Senior Software Engineer based in Costa Rica, working at Microsoft on scalable cloud systems and global policy infrastructure.

My engineering work spans migrating legacy services to Kubernetes, building change management systems, and developing internal SDKs that standardize messaging layers across the organization. I like problems where reliability at scale is the hard part.

Alongside that, I do machine learning research on medical imaging. My GECCO ’26 paper applies evolutionary optimization to decoding in vision-language models — and reports a negative result, because knowing what doesn't work is worth publishing too.

Away from the keyboard, you'll find me on hiking trails through Costa Rica's cloud forests, perfecting my pour-over coffee, or exploring the subtle world of single-origin teas.

Engineering

System Design.NETAzureKubernetesIaCCosmosDBPowerShellTypeScriptNext.js

Research & ML

PythonPyTorchVision-Language ModelsEvolutionary OptimizationExplainable AIMedical Imaging

Interests

HikingCoffeeTeaNature Photography

Research

Publications

I work on machine learning for medical imaging — specifically on understanding where vision-language models actually fail, and reporting it honestly when the answer is “this doesn't help.”

View full ORCID record
GECCO '26 CompanionPeer-reviewed · ACM

CMA-ES Decoding Optimization for MedGemma: A Flat Landscape on Chest X-Ray Reports

Michael Chen-Wang, Miguel Abreu-Cardenas, Saul Calderon-Ramirez

Proceedings of the Genetic and Evolutionary Computation Conference Companion, pp. 1441–1447 · San José, Costa Rica · 2026

  • 749 evaluations across five independent CMA-ES runs on PadChest-GR
  • Fitness landscape is flat — varying only 2.6–4.5% within each run
  • Greedy baseline holds semantic consistency at 0.97; CMA-ES drops it to 0.82
  • Locates the real bottleneck in visual perception, not text decoding
Vision-Language ModelsDecoding OptimizationCMA-ESMedGemmaChest X-RayNegative ResultPadChest
PaperCodedoi:10.1145/3795101.3814726

Projects

Things I've Built

WorthFlow

A net worth tracker built around monthly snapshots instead of transaction-level bookkeeping. Log account balances once a month and see how your finances actually trend over years — the signal most budgeting apps bury in noise.

Next.jsTypeScripttRPCPrismaTailwind CSS
Visit

CMA-ES Decoding for MedGemma

Research code behind the GECCO '26 paper: evolutionary search over temperature and top-p for a medical vision-language model, scored by a Wasserstein-distance proxy in sentence-embedding space.

PythonPyTorchCMA-ESTransformers
Source

Experience

Career Timeline

  1. Sep 2023 – Present · Costa Rica

    Senior Software Engineer

    Microsoft

    • ·Led projects to modernize legacy systems by migrating services to Kubernetes environments, resulting in improved scalability and efficiency.
    • ·Developed and implemented a change management system that significantly improved the stability and reliability of global policies.
    • ·Engineered and launched a feature to streamline user policy management, enhancing system stability, reliability, and performance.

  2. May 2021 – Sep 2023 · Costa Rica

    Software Engineer II

    Microsoft

    • ·Developed a streamlined solution for rapid updates to emergency number profiles, deprecating a legacy system with a 24-hour update cycle and providing on-call engineer tools.
    • ·Migrated a legacy service to highly secure, government-regulated cloud environments, leveraging infrastructure-as-code (IaC) for automated provisioning and standardization.
    • ·Developed a widely-adopted SDK that streamlined development workflows and enforced standardization across the organization's publish-subscribe (PubSub) messaging layer.

  3. Jul 2019 – Apr 2021 · Costa Rica

    Software Development Engineer

    Intel Corporation

    • ·Developed and launched a SaaS-based search solution in 4 months, boosting performance by 90% for over 17 million annual searches.
    • ·Researched, designed and built site-wide breadcrumbs that unifies different solutions into a single design.
    • ·Enhanced the CI/CD process for Intel's microsites by simplifying the tech stack by 30% and automating deployments and testing using Python, GitLab, Cypress, S3 Bucket, and Akamai CLI.

  4. Feb 2019 – Jul 2019 · Costa Rica

    Java Developer

    3M

    • ·Designed a new feature that allowed a better integration between the batching service and the DAM.

  5. Jun 2017 – Feb 2019 · Costa Rica

    Software Development Engineer in Tests

    Intel Corporation

    • ·Enabled the CI/CD process by automating test cases, creating frameworks and setting up the pipeline.