
Senior Applied Gen AI Engineer @PocketFM_App · prev Backend @mydukaanapp
Bangalore
Joined January 2018
- Tweets5.7K
- Following4.5K
- Followers1.1K
- Likes13.6K
Pinned Tweet
1/5
Domain experts can’t touch the code.
Engineers can’t fix what they can’t judge.
So we built Agent Doctor.
A lawyer, banker or doctor opens a run, finds the broken step, and fixes it in plain English. No agent code changes. No redeploys.
Your agent stays on your servers. Two lines to start.
उठते हुए तो बस बेचैनी है
गिरते हुए ऊँचाई का अहसास होगा
youtube.com/watch?v=VJnUpJ90…
1
105
Introducing Claude for Figma! Animate your storyboards in Figma with @genmotionhq in minutes using Claude.
Get production grade motion graphics and videos, in a fraction of time.
Abhinav Prakash retweeted
🚀 Today we're launching Agent Sandboxes in @pydantic Logfire: a profoundly different kind of sandbox.
"The world doesn't need another sandbox" they say. Maybe. Or maybe the world doesn't need another firecracker/gVisor/VM infra-fest.
Our Agent Sandboxes are different. They're just an isolated interpreter with cold start times measured in milliseconds, where the code can only call the tools you give it and use the directories you mount.
A massive bonus is you can snapshot and resume a REPL weeks later, with the snapshots costing virtually nothing.
Programmatic tool calling, solved.
Start free, $1000 of free credit to try it in the first month 👇
pydantic.io/launch-agent-san…
Abhinav Prakash retweeted
~1,100 agent threads on my server since mid october. i read one file
Friday is a @t3dotcodes thread that only coordinates: hands work to other threads, checks the proof, keeps a board. its memory lives in md files, not the chat
wrote up how it works..
harssha.com/blog/coding-setu…
Abhinav Prakash retweeted
introducing the DetectIfAI bot for X!
tag @DetectIfAI under a post with audio or video.
it'll check the audio and reply
- no AI detected
-AI detected
-partly AI, inconclusive.
Abhinav Prakash retweeted
What's impressive is that agents run within your own infrastructure, using your own data and models.
Combining trace-level debugging with plain-English fixes while keeping full control of execution is a brilliant approach. 👏
This is super cool!
Video made with @genmotionhq 🫶
4/5
What’s in it today:
• Tweak & Re-run: change the prompt, model or temperature on one step or several, re-run the whole agent on your own servers, and see what the final output becomes
• Probe: an AI debugger that bisects a bad run down to the step that broke
• Consistency Checker: replays a step 10× on the same input and shows you which steps don’t give the same answer twice
• GEPA: evolves your prompt across generations and scores each one on your own examples
Every one of them works from the dashboard. No code.
1
4
100
5/5
This is the first step. More is coming.
Built as a side project with @im_ankitjais12 .
Thanks to @souravinsights and @iNishant for testing early and telling us exactly what to improve.
If you know someone stuck between an engineer and a domain expert, send them this.
Start free: theagentdoctor.com
Launch video made with genmotion.dev by @haxzie_
Music: Beethoven, Moonlight Sonata, performed by Paul Pitman (musopen.org)
Abhinav Prakash retweeted
Came to Pondicherry to clear my head, and somehow an @Airbnb mess has consumed all my energy.
We booked a beautiful sea-view property through bnb. The pictures and videos looked great. But when we arrived, it was a completely different story.
The place was dirty, mirrors were covered in spots, nothing was properly cleaned, and there was barely any space to move around. It felt nothing like what was advertised.
We immediately called the host and explained the situation. He told us we could check out within 5 minutes and receive a full refund.
We literally checked out the next minute.
Since then, we've been going back and forth with @AirbnbHelp . We recorded the conversation with the host and shared all the evidence, but we're still waiting for a clear confirmation on our refund.
Meanwhile, we were left roaming the streets of Pondicherry late at night, trying to find another place to stay. We eventually booked another Airbnb, i think that also costs more.
@Airbnb says they're looking into it. But what exactly is there to investigate when we have the conversation, the evidence and the host's own confirmation?
We don't expect this kind of experience from a platform like @Airbnb.
A trip we planned to unwind has turned into a complete headache.
Has anyone else faced something similar with @Airbnb?
@AirbnbHelp @Airbnb_in @Airbnb @bchesky @amanpreet_bajaj
Abhinav Prakash retweeted
Hello folks,
today im launching antidraw.
an opensource product design & prototyping canvas for your coding agents powered by real code.
all my creative outputs for the past few weeks where made in antidraw, including this launch video.
i'd love for you to try and give feedback for the alpha.
Abhinav Prakash retweeted
1/ the tiny design details decide how any software feels. the motion, the states, the UI details you don't notice until they're missing.
i build a lot of personal side projects and most of my time goes into crafting those moments.. here are some of my favourites 🧵
We got neural vibe coding before gta 6 .
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
1
108















