20 yo | Founder @ Black Robotics | @fdotinc canopy | Robotics | AI
Joined January 2024
- Tweets 1,537
- Following 506
- Followers 657
- Likes 1,819
396 Photos and videos
Mandar Wagh retweeted
Jun 16
agentbrowse: Give your AI coding agent the web as a command line by @waghweb producthunt.com/products/age…
1
1
64
Jun 16
the only difference between losers and winners
is enough delusion to start from ZERO every time you FAIL !
1
3
61
Jun 15
Your coding agent can read the web. It can't use it.
One command fixes that:
> npx agentbrowse skill
Now Claude Code, Codex, Cursor, Gemini & Windsurf can drive any website from the terminal — open, click, fill, read.
Watch it work ↓
4
478
Mandar Wagh retweeted
Jun 15
Cool work, made the paper available here so more people can explore it
paperswithcode.co/paper/9836…
Jun 13
DVD-JEPA · an open-source, fully-reproducible JEPA world model
github: github.com/mandarwagh9/dvd-j…
paper : github.com/mandarwagh9/dvd-j…
demo: dvd-jepa.vercel.app/
1
10
79
8,468
Jun 15
431 downloads in 2 days is good !
Agent-browse: drive any website from the terminal and its Built for AI coding agents
"npm install -g agentbrowse"
use it in your next project !
5
435
Jun 15
People are loving fun lil experiments
Jun 13
DVD-JEPA · an open-source, fully-reproducible JEPA world model
github: github.com/mandarwagh9/dvd-j…
paper : github.com/mandarwagh9/dvd-j…
demo: dvd-jepa.vercel.app/
1
36
6,459
Jun 13
okay so i shipped so many thing in last 30 hours
i need to make it work from here to, proper get user's feedback !
the projects are too early right now, and getting feedback is harder
anyways i made it opensource, go check it out: github.com/mandarwagh9
7
364
Jun 13
A model-agnostic memory layer for robotics from egocentric human video
check this out :
github.com/mandarwagh9/egome…
2
154
Jun 13
Video streaming has 240p → 4K. It serves the same content at the quality your device network can handle.
Robot models don't have this. You ship one fixed model that's either too big for the edge or too weak for the task.
So I built adaptive-serve.
🧵[1/7]
1
3
273
Jun 13
I built a benchmark to check the router. It immediately caught a bug in the router: cloud latency was scaling with the robot's local compute.
(cpu-only "cloud" = 2100ms vs orin-nano = 105ms, same cloud model.) Fixed. That's what eval harnesses are for.
[6/7]
1
2
81
Jun 13
One model, the right call for every device and link. Big device → full model local. Small device → cloud on a good link, split on a weak link, best-effort local when the link dies.
Open source, MIT, 181 tests, green CI.
""pip install adaptive-serve""
github.com/mandarwagh9/adapt…
[7/7]
2
61