your friendly neighbourhood engineer

Joined July 2020
born in right generation huge ass neural nets cuda errors claude code iced americano
2
8
1
212
62,100
a paper i was reviewing had a citation to my work, wow i really made it
8
129
3,123
started 2023 now we are here :D
8
250
atharva ☆ retweeted
lil dashcam drives your car! hands off eyes on!!
5
4
2
68
17,345
RT @jaiselsingh: A common objection to event cameras: "what happens when nothing moves?" But you can move the camera. Every edge fires, st…
1
31
deep reinforcement learning is just about increasing the batch size obsessively starging your policy entropy
3
23
1,304
having a girlfriend is just about increasing patience and obsessively staring at her selfies
3
194
coding agents will turn everything and anything made of bits into linux shaped tools
You can really just tweak macOS however you want
17
876
in the grand scheme of things, mastering the use of computer has given me way more returns that anything ever
2
2
26
1,093
one must imagine sisyphus having llm sycophancy
1
12
795
i think its a great time to solve problems
2
17
949
atharva ☆ retweeted
meet McQueen. an autonomous RC car built to test how pretrained driving models from @comma_ai perform in the real world, using a single RGB camera and remote inference.
9
6
5
61
4,034
atharva ☆ retweeted
As our customers scale to training robot models using millions of hours of our recordings, we needed to build a new kind of data infrastructure. Terabytes of data delivered in minutes, and the capacity to deliver over a Petabyte per day to each of our customers, to multiple customers at the same time. That’s more than 10 Gigabytes a second, delivered into any cloud, wherever our customers need it. Using our customer portal, customers select precisely the data they want and it starts arriving in minutes at their destination, including AWS, GCP, Cloudflare, Hugging Face and any cloud provider who can handle the ingest speed. We’ve written about how we worked with @huggingface and @Cloudflare to build this secure, scalable infrastructure. Link in comments.
9
18
10
132
34,412
atharva ☆ retweeted
i taught a vision-language model to drive trackmania (AO1) qwen2.5-vl-3b watches the game , reads the frames, reads the car's telemetry and presses the keys. no reinforcement learning, it learns how to play based on 300 expert replays finishes about half its laps, best lap 24.6s
3
1
21
496
introducing yumi a hacky $50* umi for the YAMs. collect data quickly -> visualize -> train & deploy *camera's not included - used from the other yam lol
15
22
5
234
12,966
hacked on this fun sidequest w @vruga_ and @tirth_gada
2
15
764
more semi technical yap here xitter.cf/k7agar/status/21092928…
wrote a little blog summarizing all the robotics and intelligent systems projects i have done in past 2 years with my friends. this summarizes all the things that worked, lessons i have learned and what i think is worth pursuing ahead. link below
2
170
we envision minimalism and modularity go hand in hand hence hacky umi collection devices will lead to faster development loops. img credit: kiwi
2
10
321
more hacky data collection work here xitter.cf/k7agar/status/21050413…
hacky leader arm 1/10th the cost of og leader arm
9
286