Software engineer · systems builder · Physical AI explorer

Kai Kyungsuk Song I build things that have to work.

Senior Software Engineer at Microsoft in the Los Angeles area, with 20+ years across software, distributed systems, cloud data platforms, advertising technology, and now robotics and Physical AI. I like hard systems problems, useful documentation, and experiments that leave evidence behind.

LinkedIn profile GitHub Email me

About

Longtime software engineer. Newer robot mechanic.

My career has moved through web platforms, advertising technology, large-scale data systems, forecasting, cloud infrastructure, and applied AI. Physical AI is the newest chapter: taking decades of software instincts into a world where wiring, batteries, motors, sensors, and gravity all get a vote.

20+ yearsProfessional software engineering
MicrosoftSenior Software Engineer
W3CImmersive Web API reviewer
PatentInventor on dynamic content relocation

Selected engineering

Systems, platforms, and a few interesting detours.

This is a landing page, not a 20-year changelog. LinkedIn has the longer version.

Current

Forecasting & large-scale data systems

Engineering forecasting and monetization systems at Microsoft, with the usual distributed-systems ingredients: APIs, data pipelines, performance, reliability, cloud infrastructure, and operational tradeoffs.

See current work on LinkedIn →
Open web

W3C Immersive Web

Volunteer API reviewer for the Immersive Web Community Group, connecting practical engineering with open web standards.

Patent

Dynamic relocation of graphical digital content

Inventor on techniques for dynamically repositioning digital content based on changing client-side conditions.

Earlier chapters

Ad tech, analytics, video & Apple TV

Selected projects on my LinkedIn include Monetize Insights, an AppNexus Apple TV SDK, and a video framework—different surfaces, same obsession with making the system actually work.

Engineering journals

I publish the learning process, not just the victory lap.

Notes are organized around decisions, experiments, failures, architecture changes, measurements, and lessons worth preserving.

Physical AI · Robotics · 2026 →

Engineering Physical AI Journal

A chronological record of moving from software systems into robotics: choosing a platform, preparing ROS 2 and GPU environments, testing real hardware, preserving factory baselines, improving remote development, and discovering that even charging can become an engineering project.

Preserve the timestamp. Preserve the evidence. Publish the lesson.
Read the journal 한국어 저널 More engineering on GitHub

A few less-formal facts

Because engineers are allowed to have a personality.

  • After 20+ years of making distributed systems behave, I decided robots looked easier. That assumption is currently under active review.
  • I document failures because future-me is a surprisingly unreliable incident historian.
  • My current definition of Physical AI: software that can finally bump into the furniture.

Connect

Find the engineering work and technical conversation.

LinkedIn is the professional timeline. GitHub is where the artifacts live. X carries shorter technical updates and conversation.