JaehoonSon
I turn open research questions into working systems—evaluating new models, building agents, and shipping tools for real users.
Writing
Five AI models tried to make the rainbow in one cup
Claude, Gemini, GPT, Grok and DeepSeek each ran a real pump rig on their own, scored by what a webcam measured in the cup.
Education
B.A. in Computer Science + Mathematics
GPA 3.83/4.0. Three-time YC and two-time DeepMind hackathon winner; member of Swarthmore Competitive Programming.
Experience
Software Engineer Intern
Owned reliability and adoption for WorkLoop, which runs long-running and scheduled LLM agents in isolated per-team sandboxes, and built Tesla's first internal benchmark for agent context compaction.
Undergraduate AI Researcher
Built a self-improving agent pipeline that exceeded LoRA fine-tuning across three research workflows, alongside evaluation systems that improved performance by 40%.
Software Engineer Intern
Shipped LLM-assisted industrial data tools used by plant teams and cut average data retrieval time by more than 50%.
Software Engineer Intern
Built a content-rendering pipeline for 100,000+ users and helped deliver 30% faster load times with 50% less server usage.
Software Engineer Intern
Helped launch a credit card service, strengthen access controls, and increase customer onboarding by 15%.
Projects
Hermes
An AI customer intelligence platform that has sent 80,000+ personalized emails across coordinated outreach channels.
SlideBud
An agentic presentation platform with grounded research workflows, built for more than 10,000 users.
MakeYourWebsite
An AI website builder used by 8,000+ people to publish more than 900 live websites.