Full-stack Developer @monkfunnel. Building products and shipping things. Sharing what I learn about code and startups. DMs open.
Earth
Joined February 2024
- Tweets441
- Following433
- Followers47
- Likes712
the one thing i am realizing about software eng:
Writing code is only one part of building software,
Understanding the problem, making architectural decisions, debugging unexpected behaviour, and also knowing why something works are just as important
PostgreSQL stores data in 8kb pages
Not rows or individual records.
So now i am trying to understand why,
This database rabbit hole is getting ridiculous
been thinking about this lately:
AI is making coding way faster, so I guess the bigger skill now is knowing what to build, how it should fit together, and how to judge what AI gives you.
Finished Chapter 2 of Designing Data-Intensive Applications.
Biggest takeaway: "documents vs relational" isn't old vs new, it's a tradeoff that's been resurfacing under different names since the 60s.
link medium.com/@ankitmmehta08/wh…
rebuilding an AI coding agent from scratch and the annoying realization is that AI is like 20% of the problem. the rest is just... understanding why streaming works the way it does, how tool execution is wired up, why the terminal UI is built like that, how state gets managed. turns out AI is just a layer sitting on top of normal software engineering that people skip past.
React doesn't render to a browser. It renders to whatever you tell it to.
building a terminal AI agent with OpenTUI, same useState, same useEffect, same JSX, but the output is terminal escape codes, not DOM nodes.
<box> is just <div> for your terminal.
The 2nd edition of 'Designing Data Intensive Applications' dropped and i finally sat down with chapter 1.
I thought i knew 'data' pretty well, but it turns out i was only seeing the UI layer version of it.
And the one section I didn't expect in a systems book: data protection laws (GDPR, India's DPDP, CCPA) and AI bias.
Engineers now have to design for "right to be forgotten" in append-only systems. That's a genuinely hard engineering problem, not just a legal checkbox.
for years, i thought YouTube live meant my browser was receiving a continuous stream from the creator.
turns out it's mostly downloading tiny video files created just seconds earlier.
sometimes "live" is just a really fast VOD pipeline.
stages of debugging:
1. this should work
2. why doesn't this work
3. how is this working
4. don't touch it
backend concept that finally clicked this week: indexing.
i knew "add an index to make queries faster."
i did not actually understand WHY until i watched a query
go from 800ms to 4ms on a table I'd populated with fake rows.
an index is just a sorted lookup structure so the database
stops scanning every single row to find what you asked for.
like the index at the back of a book vs reading the whole thing.
the catch nobody mentions: every index you add makes
writes slower, because now there's more to keep updated.
everything in backend is a tradeoff, nothing is free