Jev vs LLMs, Thinking Fast and Slow
Why AI agents should hand their small, routine decisions to fast, lightweight models instead of a big LLM.
// hello, world. i’m subin s vI turn
I turn
chaos
into systems.
Every project starts as chaos. Ten ideas, three deadlines, one whiteboard.
I’m a Principal Architect in Chennai, and my job is turning that into something that still works on a Friday evening. Lately, that something usually has AI inside.
~ $ neofetch
subin@chennai
─────────────
roleprincipal architect
focusagentic ai · llm evals · voice
shipsmobile · desktop · web
shipped8 side projects, 1 watch face
blogJev vs LLMs, Thinking Fast and Slow read ↗
homechennai, india
type help, or tap a command below.
01 / side projects
Shipped, not just started.
Apps and dev tools I’ve built and published on Google Play, the Microsoft Store and the web.
02 / blog
All postsFresh from the blog.
latest post
01 Dec 2019Programmers Health Guide
16 Nov 2019Part 3 [touch & mkdir] - Know your commands
27 Oct 2019Part 2 [echo & ls] - Know your commands
19 Oct 2019Part 1 [pwd & cd] - Know your commands
03 / stack
Open in my editor.
Who I am, how I work and what I use. Two files, click around.
EXPLORER
// the short version
val subin = Engineer(
role = "Principal Architect",
base = "Chennai, India",
ships = listOf("Mobile", "Desktop", "Web"), // mostly Flutter
now = listOf("System design", "Agents", "LLM evals", "Voice AI"),
)
// how every new project starts
fun Engineer.onNewProject() {
sketchTheBoxes()
askWhatHappensWhenItFails()
writeTheDecisionDown() // future me will ask why
buildTheSmallestThingThatWorks()
measureItBeforeCallingItDone()
}
⎇ main · ✓ 0 problemsKotlin · UTF 8 · LF
04 / contactGot chaos?
Got chaos?
Send it over.
A system to design, an AI idea that needs a second opinion, or a hello about something you read on the blog.