@mina
InitiateMina Choi
Research Ranger Lv.19 · Reader of 400 Tabs
Grad student who uses models to read faster, not to think less. Keeps a running argument with every paper she cites.
Build guide · updated Sep 6, 2026
A literature pipeline: nightly arXiv digest, adversarial summaries on demand, and a citation database you can actually trust.
- Steps
- 5 (1 optional)
- Time
- 1 h 50 min
- One-time
- $1,398
- Monthly
- $20/mo
Every slot lists where to get it and what it costs. The full walkthrough is under Follow this build.
- Main modelEpic
Gemini
Google
Whole PDFs in one context
- Coding toolCommon
VS Code
Microsoft
Notebooks and plots
- VoiceRare
Whisper
OpenAI
Transcribing seminar recordings
- SearchEpic
Perplexity
Starting point, never the ending
Free/mo (Free tier) · Free tier
Site:Perplexity (opens in a new tab) - AutomationCommon
cron / launchd
Nightly arXiv digest
- HardwareCommon
MacBook
Apple
Library laptop
- WearableRare
Apple Watch Ultra
Apple
Timers for reading sprints
- Notes & tasksRare
Notion
Lit review database
Free/mo (Education plan) · Education plan
Site:Notion (opens in a new tab)
- Research
- 91
- Writing
- 78
- Coding
- 52
- Automation
- 38
- Tab discipline
- 5
- Adversarial summary
- Citation hygiene
- Reading sprint
- Notebook plotting
- Closing tabs (locked)
I read the abstract myself, then ask the model to argue against the paper. If it can't find a weak point, I read the paper properly. If it can, I check whether the weak point is real.
- A nightly cron job pulls new arXiv listings for my keywords into Notion.
- Seminar audio goes through Whisper, then a summary lands next to my own notes.
- Every claim I cite gets a one-line "why I believe this" next to it.
The model reads faster than me. It does not read better than me. Yet.
- Ask the model to argue against the paper before you read it119
An adversarial summary tells you where to look. A neutral summary tells you nothing.
Research
- Never cite what the model told you. Cite what you checked.88
Models invent references. Keep a 'why I believe this' line next to every citation.
Research
- A nightly arXiv digest with a cron job and a model67
Pull new listings for your keywords, have the model rank them, wake up to five links instead of fifty.
Research
- Read in 25-minute sprints with a wrist timer, summarise in the gap39
Timer on the Watch, one paper per sprint, a spoken summary in the five minutes after.
Daily life