adish pandya
5 posts


Now that our 15 member llm team is infamous, time to expand for next time!
If you have done one or more of the following, then please reach out.
- pretrained a model of any size, from scratch
- posttrained any base model, end to end (data curation, sft, rl)
- are a pytorch wizard
- are a cuda kernel master
- you have any other relevant skills and work to back it up
firstnamesarvamai
English

I pulled it off finally!
I had this huge goal for the project Mandy, to add Workflows. Most big players in the industry for API testing NEVER thought of workflows and chaining multiple requests to automate and get valuable data.
Now its here at 0.3.0
github.com/Rahuletto/mand…



English
adish pandya retweetledi

happy to announce that our paper from AI4Bharat has been accepted to the icbinb workshop at ICLR 2026 🎊
work done with @prajdabre @AdishPandya

English

short story how this paper came about
in the poster session of AACL 2025 (Dec 21 2025), i noticed a paper that found a flaw in LLMs in a particular task. it tested in only one cultural nuance. i wondered would it translate to other cultural cases. one night of experiments on the night of 22nd, IT ACTUALLY TRANSLATED TO THE OTHER CASES. it was a cool observation and i just informed @prajdabre that night.
this was not going to be a paper.
fast forward to 20th Jan 2026, I saw this workshop ICBINB that especially wanted to get papers which detected flaws in LLMs, so that night I got up and started working on this. We needed to expand to more tasks, some more ablations + some rmore cultural nuances needed to be addressed. the work was greatly helped by @AdishPandya as well. and we finished writing this paper + running the remaining 30%ish work in these few last days.
Really fun working in environments like this :) and I really feel this is the reason I love working in research!
Preprint will be out soon :P
neural nets.@cneuralnetwork
one more paper done this was a draft for a long time, finished it up and submitted
English

