Day 2: Accelerating - Samuel Cheng

Second day in a row building stuff until past 12am, this is addicting for my ADHD brain. My CS background and time spent learning web dev years ago are definitely helping me speed run all of this - it would be daunting without.

I am not even on the agentic side of things yet, just building manually.

Learning about usage limits of Codex
- Understand why people hit usage limits
Optimizing through prompting first in Chatgpt
Annotated aspects in Codex’s in-app website but it didn’t work too well
Building like a couple of static websites, a new revamped personal website that connects to this Notion site, and my deep research dashboard at the same time
Opened a project in Cursor for the first time
Set up my Codex pet, but I really don’t look at it
Deep tech research dashboard
- Added light mode
- Refactored UI from a vertical scroll to a responsive dash
- Refactored content of UI to focus on a ‘world class’ score of a company, where the tech sits in the value chain, and what the tech actually is
- Too tired so here is what i did with that dashboard again: I built the first version of the “Class Potential” engine. Instead of just summarizing companies, the app now tries to evaluate whether a company could become world-class, regional-leading, niche-leading, or too early to call. I refactored that logic several times: first as a weighted scorecard, then into a more investment-style factor model covering global category rank, technology differentiation, founder strength, market importance, competitive position, funding quality, traction, and strategic relevance. A key lesson was that scoring is hard. SpaceX initially came out as merely “Regional Leader Potential,” which exposed that the logic was too fuzzy and too dependent on missing public data. I changed the model so missing information lowers confidence rather than punishing potential, added override rules for globally dominant companies, and tested it against world-class and smaller-company examples. By the end of the night, the product had the bones of a real deep-tech research system: search, AI research, saved profiles, and a differentiated scoring framework aimed at answering the question investors actually care about: “Could this company become world-class?”
Personal Website Progress
- literally generated the entire thing i think?
- got real time fetching, should update without touching vercel per 1-5min
- created a notion connector
- had trouble with database id, still don’t know how i fixed this
- blogs split into 3 categories: essay, build log, and notes
Reviewing how to set up git repos with gitignore of all the local env vars
Using warp primarily to do git status, add, commit, and push, which i don’t think is really needed probably just Cursor CLI right? But coding this stuff in natural language is easier
I just upgraded to ChatGPT Pro lol, it’s definitely worth it
- thought that i hit my rate limit though I was at 80% and I had to restart it, it fixed it
literally so tired it is 1:34 am i have not been so locked in
deployed on vercel twice - this blog and also the dashboard at https://deeptech-research-os.vercel.app/
ideas for features for next night for the company dashboard
- click on score component to dig deeper
- score update and graph to track
- export into PDF report feature
next features for personal website
- integrate project links → how to deploy my projects faster
- clean up first page ui
- link my personal links

Work

did most of my work through claude cowork
got it to help structure my writing