Jinseop Song

About Me

Hi! I'm Jinseop (or David) Song, a high school senior (class of 2026).

I study time series domain, Agentic AI, and Astrophysics.
My focus is on token permutation methods (mostly attention variants) in time series forecasthing models and reinforcement learning-based method (full finetuning for Tool) to increase tool accessability of LLMs.

My dream is to create a model that predicts the world. Whether that's predicting "Who's winner of 2024 election in Korea" or "The price of Big-mac in 2030".. Yes.. The world. The universe on macroscopic scale has patterns (so-called history), and on microscopic scale, there are randomness (so-called entropy).
I believe a good combination of time series forecasting model + Physics + Multi-Agent LLM can acheive this in near future.

My Journey

I began programming in 2016, back in third grade, using Sketch (block coding). Two years later, I started learning Python but soon realized it was too difficult for me. (I was also stuck at learning programming as my English wasn't good, and Python was all English...)

By 7th grade, I had become a full-stack developer and have been one for over five years now. I primarily use Flutter (with BloC or Riverpod) for app development and rely on Supabase, Firebase, and GCP for backend services.

Over the years, I've built some private apps like a personal MacBook version of YouTube Music (using yt-dlp + Flutter) and my school (jeju) 's internal chat app (Firebase) to my recent project, Mealzy. I've also worked as a frontend developer at startups to learn how to collaborate with designers and backend developers.

In 8th grade, I tried Python again, and one year later I started to learn AI, motivated by ChatGPT outbreak. Then I found myself in time series forecasting, and I've been reading papers and on open source communities to understand why/how people love AI.

You can find my CV here.