The Context Window
50 engineers. Production AI. India. Invite-only.
Not a newsletter. Not a Discord for sharing papers nobody reads.
Every Monday morning someone posts what broke in their production system last week. Not a write-up. Not lessons learned. The actual failure, the debugging session, the thing no documentation prepared them for.
We are building the founding cohort now. Ten people who set the culture for everyone who comes after.
// admission
// what actually happens here
#what-broke — every monday
One post per week. What broke, how you found it, what it took to fix it. This is where the knowledge that never makes it into blog posts actually lives.
#deep-dive — monthly
One member walks through a hard problem they are working on. 20 minutes, then the room picks it apart. No slides required. Closest thing to free consulting with people who have done the same work.
#compensation — anonymous
We are building the most honest AI engineer compensation database in India. Member-only, anonymous, aggregated. The number that does not exist anywhere publicly.
// the kinds of conversations
Our retrieval precision was 0.71 in staging and 0.43 in prod. Three hours of debugging. Here is why.
Replaced LangChain with direct API calls. 40% less code. Was it worth the migration pain?
What does ML model monitoring actually look like at a company that is not Big Tech?
We re-finetuned the same base model four times in six weeks. Here is what we got wrong each time.
Staff AI engineer at a Series B in Bangalore. Total comp breakdown, honest.
// tools we actually talk about
// founding cohort
First 10 members set the culture. Applications are reviewed manually.
Not everyone gets in.
apply →