Thanmay

MS Student @WSAI_IITM | @iitmadras | @AI4Bharat | @teamIvlabs

[Thanmay @yamnaht]

@AlhamFikri Waduh Prof 😂 plis dihapus wkwkwk jgn kasih ide

[Manuel @Manuel_NLP]

Important for anyone who did not get into EMNLP. You can still commit to AACL 🔽

[Thanmay @yamnaht]

Need this confidence in life

[Amit LeVi @AmitLeViAI]

See you at EMNLP!

We have several papers there! I’ll be there for a short visit, so feel free to message me.

#EMNLP

[Thanmay @yamnaht]

Reviewer when instructors do not have experience in the proposed topic: "Limited expertise"

Reviewer when instructors have experience in the proposed topic: "Self-invited Talk"

What are we supposed to do? Have expertise, but not enough expertise to be suspicious?

[Raj Dabre @prajdabre]

I got the same kind of review. Not only does the reviewer discriminate based on experience but also based on institution.

[Thanmay @yamnaht]

Thank you for the incredible engagement at the recent
@WSAI_IITM Annual Research Showcase. As promised, you can access the app right here. iOS support coming soon!

Drop your feedback in the replies so we can keep improving.

[AI4Bharat @ai4bharat]

🚀 Excited to release IndicSwipe! 🇮🇳 The first fully open-source gesture-typing keyboard for 22 Indian languages. It supports multiple native scripts and comes with our custom-trained neural swipe models!

This is our first iteration, and many improvements will follow.

📱 Play Store: https://t.co/V80K3SuU0J

💻 GitHub: https://t.co/1iLZC9YTKO

More features coming soon!

📝 Share your feedback: https://t.co/a1S6bPeiDA

@srihari_118 @yamnaht @prajdabre @MiteshKhapra @ai4bharat @WSAI_IITM @iitmadras

#IndicSwipe #OpenSource #AndroidDevelopment #AndroidDev #MachineLearning #Kotlin #IndicLanguages #AI4Bharat #MobileAppDevelopment #TechForIndia #OpenInnovation

[Thanmay @yamnaht]

TIL Jakarta has the unfortunate title of the world’s fastest sinking city

i thought metaphorically “sinking” at first and then i googled it💀

[anny (formerly Midas) @MAD_as_25]

@yamnaht Coz it's sinking?

[Thanmay @yamnaht]

TIL Indonesia’s capital is no longer Jakarta.

hah kok gitu sih

[Thanmay @yamnaht]

Verifiable Instruction Following is a low-cost soln for Indic langs, esp given the lack of reliable LLM-as-a-Judge or human annotators

Excited to see its use in the Indic RL ecosystem

Glad our benchmark could help!

Do check out @adityapuranik99's post

Open source at its best.

[Aditya Puranik @adityapuranik99]

https://t.co/6RMKbZwSZp

[Chinmay @ChinmayKak]

Introducing heiretsu

A minimal, from-scratch implementation of 4D parallelism (DP/TP/PP/EP) in pure PyTorch, with support for both dense and MoE training.
Code and sample W&B runs below!
Please do check it out, and RT/QT for the reach:)

[Raj Dabre @prajdabre]

Transliteration is a seriously underrated technique for improving NLP performance in low-resource settings, especially for Indian languages. We put together a survey highlighting the various strategies people have used over the years. Please enjoy!

Link in comments.
cc @cneuralnetwork @yamnaht (Pls tag relevant people)

[Rishika Bhagwatkar @rishika2110]

Excited to share that I'll be at #NeurIPS2025 from Dec 2-7, presenting two of my works 🚀

1. LLM Agent Safety Against Prompt Injections
2. Anomaly Detection Using VLMs

I’m looking for PhD positions starting Fall 2026.

Happy to connect, feel free to DM :)

Poster details in 🧵

[Thanmay @yamnaht]

@HarveenChadha We had done a survey of this in our tutorial
https://t.co/D2w14q4UVv and you can find the relevant studies in the reading list (references 30-41) and slides.

[Thanmay @yamnaht]

In general, studies show that the reasoning performance in a language is attributed to the proportion of that language's data during training. This is true for both instructing the model to reason in a language (forced) and letting it reason and see what language it uses (non-forced), especially math tasks. In this case, Chinese is predominant in the training and also more token-efficient as others have pointed, and maybe that's why it switches to Chinese (non-forced) because of length penalties during RL training. Even so, these models are also trained with language-consistency rewards, so it would be interesting to understand the interplay between language-consistency rewards and length penalties during RL training, specifically in a multilingual setting, and whether token-efficiency overshadows language-consistency in such cases.

[Thanmay @yamnaht]

@prajdabre Interested

[Thanmay @yamnaht]

@cneuralnetwork Ufff chorom🔥

[Thanmay @yamnaht]

@dongxi_nlp 天哪，已经三年了吗

[Thanmay @yamnaht]

@AlhamFikri dulu gw bersekolah di SEA, tp sekarang di India. bisa apply gak ya?

[Thanmay @yamnaht]

@cneuralnetwork Go for it. There are already a lot of courses, so think how you'd like to differentiate yours.

[Chinmay @ChinmayKak]

Super delighted to share that our paper on LLMs and their forecasting ability has been accpeted at the AIR-FM workshop @RealAAAI 2026 in singapore!
Research done @lossfunk by me and @paraschopra .
Preprint soon:)

[Thanmay @yamnaht]

If you’re at #EMNLP25 and are interested in multilingual LLMs, do join us for our tutorial.

We’ll keep it broad, intuitive, and easy to follow for everyone. Perfect if you’ve ever wondered how LLMs go multilingual!

[AI4Bharat @ai4bharat]

We’re excited to announce that @ai4bharat is leading a tutorial at #EMNLP25 🎉

🗣️ “Data and Model-Centric Approaches for Expanding LLMs to New Languages”

🗓️ Nov 8 | 14:00–17:30 CST
📍 Suzhou, China

Presenters: @anoopk, @prajdabre, @RudraMurthyV, @SafiKhan2k & @yamnaht

We’ll explore how to expand LLMs beyond English, with data and model-centric strategies for low-resource languages 🌍

If you’re at EMNLP, join us live 👋
📍 Room A108 | 14:00 CST
🔗 Learn more: https://t.co/eUqsabWIOy

@WSAI_IITM @iitmadras

#AI4Bharat #MultilingualAI #LLMs #LanguageTechnology #NLP #LowResourceLanguages