Apparently my clg had hired a training agency to take mock interviews
i just attended that
i was tasked with designing a parking lot system
the interviewer was satisfied with my solution even though i have never listened or solved any sys design questions like this
20. Still in college. Already shipped:
- Self-healing MLOps. No humans needed.
- Voice agents at sub-900ms for 50+ clients.
- 20+ AI agents live across 3 industries.
- OSS contributor.
I haven't even graduated yet.
Founders in AI - DMs open.
OMG, lovable.com is literally redirecting to a lingerie site right now 😭
This is the same Lovable that just crossed $400M+ in annual recurring revenue… and they don’t even control their main domain?
That’s actually insane
AMD Senior AI Director confirms Claude has been nerfed. She analyzed Claude's session logs from Janurary to March:
> median thinking dropped from ~2,200 to ~600 chars
> API requests went up 80x from Feb to Mar. less thinking and failed attempts meaning more retries, burning more tokens, and spending more on tokens
> reads-per-edit dropped from 6.6x → 2.0x. model stops researching code before touching it.
> model tried to bail out or ask "should i continue" 173 times in 17 days (0 times before March 8).
> self-contradiction in reasoning ("oh wait, actually...") tripled.
> conventions like CLAUDE.md get ignored because there's less thinking budget to cross-check edits
> 5pm and 7pm PST are the worst hours, late night is significantly better. this means the thinking allocation is most likely GPU-load-sensitive.
this issue in llms can't be fixed
using more gpus,
pruning,
or even quantization. because the problem isn’t compute.
this is where the concept of "speculative decoding" comes in. read to know more.
linkedin.com/posts/ajitashw…
google big code : qualifier round (review)
mcqs were good (mostly)
straightforward aptitude, moderate oops concept and easy (for me) ml, gen ai questions
the programming one: i done it using binary search with greedy
built an event-driven task processing backend in go. worker pool with shutdown, postgres with jsonb payloads, dual storage strategy, structured slog logging, and a rest api service.
github: github.com/ajitashwath/ta…
everyone talks about founders
nobody talks about the team
so here i am talking about the people who build matiks.
first up, ritesh
he was a 3rd year student interning somewhere for ₹25k
we offered ₹5k
he still joined
not for money but for learning. said working closely with IIT grads would teach him more
joined in september'24, the same month we had all quit our jobs
we moved to bangalore in october
he came too, on that stipend
we were working 7 days a week then
told him to take sundays off and he said,
“what would i even do?”
over time he’s become like a younger brother
still an intern, graduating this year
but i’ll bet ₹5L he’s better than most engineers with 5 years of experience in bangalore.
rare to find people like @k123ritesh
built a smart agriculture monitoring system
simulated the entire irrigation pipeline:
sensors → mqtt → pi gateway → fastapi cloud → ml model → decision engine → pump controller → dashboard
ml predicts irrigation needs while rule-based guards prevent unsafe watering.
Disgusting scene ever. Pedophilic ra** scene was conceived by a director who couldn’t think beyond - only for the hero fans to jerk off to the hero who goes on a killing rampage. Disgust is the word