Thanmay (@yamnaht) header
Thanmay (@yamnaht) profile picture
Joined September 2021
Photo by Thanmay (@yamnaht)Photo by Thanmay (@yamnaht)
Thanmay retweeted
๐Ÿš€ Excited to share that we're launching IndicSwipe Mark I today! Type in 22 Indian languages with gesture swipe, voice typing, and translation all in one keyboard. Give it a try and let us know what you think. Your feedback means a lot to us! ๐Ÿ™
๐Ÿš€ ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—ฆ๐˜„๐—ถ๐—ฝ๐—ฒ ๐—ท๐˜‚๐˜€๐˜ ๐—ด๐—ผ๐˜ ๐—ฎ ๐—ฏ๐—ถ๐—ด ๐˜‚๐—ฝ๐—ด๐—ฟ๐—ฎ๐—ฑ๐—ฒ! ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—ฆ๐˜„๐—ถ๐—ฝ๐—ฒ ๐— ๐—ฎ๐—ฟ๐—ธ ๐—œ ๐—ถ๐˜€ ๐—ต๐—ฒ๐—ฟ๐—ฒ. ๐ŸŽ‰ A few months ago, we launched IndicSwipe, a free, on-device gesture-swipe keyboard for 22 Indian languages. Thank you for the downloads, the reviews and the feature requests. You told us what you wanted, and we built it. ๐—ง๐—ผ๐—ฑ๐—ฎ๐˜†, ๐˜„๐—ฒ'๐—ฟ๐—ฒ ๐—น๐—ฎ๐˜‚๐—ป๐—ฐ๐—ต๐—ถ๐—ป๐—ด ๐˜๐—ต๐—ฒ ๐—ป๐—ฒ๐˜„ ๐˜ƒ๐—ฒ๐—ฟ๐˜€๐—ถ๐—ผ๐—ป ๐—ผ๐—ป ๐˜๐—ต๐—ฒ ๐—ฃ๐—น๐—ฎ๐˜† ๐—ฆ๐˜๐—ผ๐—ฟ๐—ฒ! ๐Ÿ‘‡ ๐ŸŒ๐—ง๐—ฟ๐—ฎ๐—ป๐˜€๐—น๐—ฎ๐˜๐—ฒ ๐—ฎ๐˜€ ๐˜†๐—ผ๐˜‚ ๐˜๐˜†๐—ฝ๐—ฒ, ๐—ฝ๐—ผ๐˜„๐—ฒ๐—ฟ๐—ฒ๐—ฑ ๐—ฏ๐˜† ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—ง๐—ฟ๐—ฎ๐—ป๐˜€๐Ÿฎ Write in one language and send in another. Translate between English and 22 Indian languages, and between Indian languages, without leaving your keyboard. ๐ŸŽ™๏ธ๐—ฉ๐—ผ๐—ถ๐—ฐ๐—ฒ ๐˜๐˜†๐—ฝ๐—ถ๐—ป๐—ด, ๐—ฝ๐—ผ๐˜„๐—ฒ๐—ฟ๐—ฒ๐—ฑ ๐—ฏ๐˜† ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—–๐—ผ๐—ป๐—ณ๐—ผ๐—ฟ๐—บ๐—ฒ๐—ฟ Speak naturally in your language and watch it turn into text. Our speech models are trained on Indian speech, accents and code-mixing. ๐Ÿ”ค๐—ก๐—ฎ๐˜๐—ถ๐˜ƒ๐—ฒ ๐˜€๐—ฐ๐—ฟ๐—ถ๐—ฝ๐˜ ๐˜๐˜†๐—ฝ๐—ถ๐—ป๐—ด Type directly in your language's script, with no transliteration needed. โœ๏ธ๐—ค๐˜‚๐—ถ๐—ฐ๐—ธ๐—ฒ๐—ฟ ๐—ฐ๐—ต๐—ฎ๐—ฟ๐—ฎ๐—ฐ๐˜๐—ฒ๐—ฟ ๐—ฒ๐—ฑ๐—ถ๐˜๐—ถ๐—ป๐—ด Fix a matra, swap a conjunct or correct a single character in fewer taps. ๐ŸŽจ๐—ฃ๐—ฒ๐—ฟ๐˜€๐—ผ๐—ป๐—ฎ๐—น๐—ถ๐˜‡๐—ฒ๐—ฑ ๐—ธ๐—ฒ๐˜†๐—ฏ๐—ผ๐—ฎ๐—ฟ๐—ฑ ๐—น๐—ฎ๐˜†๐—ผ๐˜‚๐˜๐˜€ Arrange the keyboard the way you like to type. โšก๐—™๐—ฎ๐˜€๐˜๐—ฒ๐—ฟ ๐—ฒ๐˜ƒ๐—ฒ๐—ฟ๐˜†๐˜„๐—ต๐—ฒ๐—ฟ๐—ฒ Under the hood, we've made swipe, predictions and the whole app noticeably faster. โœ๏ธAnd of course, gesture swipe is still at the core, across multiple scripts, with Roman output for Hinglish, Tanglish and more. ๐Ÿ“ฒ๐——๐—ผ๐˜„๐—ป๐—น๐—ผ๐—ฎ๐—ฑ ๐—ผ๐—ป ๐—”๐—ป๐—ฑ๐—ฟ๐—ผ๐—ถ๐—ฑ: play.google.com/store/apps/dโ€ฆ ๐ŸŽ๐—ถ๐—ข๐—ฆ ๐˜‚๐˜€๐—ฒ๐—ฟ๐˜€, ๐˜†๐—ผ๐˜‚'๐—ฟ๐—ฒ ๐—ป๐—ฒ๐˜…๐˜! IndicSwipe for iPhone is coming in a few weeks. Stay tuned. This is still early, so tell us what you think in the comments or in a Play Store review. ๐—ฌ๐—ผ๐˜‚๐—ฟ ๐—ณ๐—ฒ๐—ฒ๐—ฑ๐—ฏ๐—ฎ๐—ฐ๐—ธ ๐˜€๐—ต๐—ฎ๐—ฝ๐—ฒ๐˜€ ๐˜„๐—ต๐—ฎ๐˜ ๐˜„๐—ฒ ๐—ฏ๐˜‚๐—ถ๐—น๐—ฑ ๐—ป๐—ฒ๐˜…๐˜. ๐Ÿ™ ๐—•๐˜‚๐—ถ๐—น๐˜ ๐˜„๐—ถ๐˜๐—ต โค๏ธ ๐—ฏ๐˜† @srihari_118 @yamnaht @akkiisfrommars @n_narasim456 @prajdabre @MiteshKhapra @anoopk @manickam24 Saranya Sukumaran @ai4bharat @WSAI_IITM @iitmadras #IndicSwipe #AI4Bharat #IndicTrans2 #IndicConformer #IndicXlit #IndicLanguages #MadeInIndia
1
1
2
135
IndicSwipe just got a big upgrade! Drop your feedback in the replies so we can keep improving.
๐Ÿš€ ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—ฆ๐˜„๐—ถ๐—ฝ๐—ฒ ๐—ท๐˜‚๐˜€๐˜ ๐—ด๐—ผ๐˜ ๐—ฎ ๐—ฏ๐—ถ๐—ด ๐˜‚๐—ฝ๐—ด๐—ฟ๐—ฎ๐—ฑ๐—ฒ! ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—ฆ๐˜„๐—ถ๐—ฝ๐—ฒ ๐— ๐—ฎ๐—ฟ๐—ธ ๐—œ ๐—ถ๐˜€ ๐—ต๐—ฒ๐—ฟ๐—ฒ. ๐ŸŽ‰ A few months ago, we launched IndicSwipe, a free, on-device gesture-swipe keyboard for 22 Indian languages. Thank you for the downloads, the reviews and the feature requests. You told us what you wanted, and we built it. ๐—ง๐—ผ๐—ฑ๐—ฎ๐˜†, ๐˜„๐—ฒ'๐—ฟ๐—ฒ ๐—น๐—ฎ๐˜‚๐—ป๐—ฐ๐—ต๐—ถ๐—ป๐—ด ๐˜๐—ต๐—ฒ ๐—ป๐—ฒ๐˜„ ๐˜ƒ๐—ฒ๐—ฟ๐˜€๐—ถ๐—ผ๐—ป ๐—ผ๐—ป ๐˜๐—ต๐—ฒ ๐—ฃ๐—น๐—ฎ๐˜† ๐—ฆ๐˜๐—ผ๐—ฟ๐—ฒ! ๐Ÿ‘‡ ๐ŸŒ๐—ง๐—ฟ๐—ฎ๐—ป๐˜€๐—น๐—ฎ๐˜๐—ฒ ๐—ฎ๐˜€ ๐˜†๐—ผ๐˜‚ ๐˜๐˜†๐—ฝ๐—ฒ, ๐—ฝ๐—ผ๐˜„๐—ฒ๐—ฟ๐—ฒ๐—ฑ ๐—ฏ๐˜† ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—ง๐—ฟ๐—ฎ๐—ป๐˜€๐Ÿฎ Write in one language and send in another. Translate between English and 22 Indian languages, and between Indian languages, without leaving your keyboard. ๐ŸŽ™๏ธ๐—ฉ๐—ผ๐—ถ๐—ฐ๐—ฒ ๐˜๐˜†๐—ฝ๐—ถ๐—ป๐—ด, ๐—ฝ๐—ผ๐˜„๐—ฒ๐—ฟ๐—ฒ๐—ฑ ๐—ฏ๐˜† ๐—œ๐—ป๐—ฑ๐—ถ๐—ฐ๐—–๐—ผ๐—ป๐—ณ๐—ผ๐—ฟ๐—บ๐—ฒ๐—ฟ Speak naturally in your language and watch it turn into text. Our speech models are trained on Indian speech, accents and code-mixing. ๐Ÿ”ค๐—ก๐—ฎ๐˜๐—ถ๐˜ƒ๐—ฒ ๐˜€๐—ฐ๐—ฟ๐—ถ๐—ฝ๐˜ ๐˜๐˜†๐—ฝ๐—ถ๐—ป๐—ด Type directly in your language's script, with no transliteration needed. โœ๏ธ๐—ค๐˜‚๐—ถ๐—ฐ๐—ธ๐—ฒ๐—ฟ ๐—ฐ๐—ต๐—ฎ๐—ฟ๐—ฎ๐—ฐ๐˜๐—ฒ๐—ฟ ๐—ฒ๐—ฑ๐—ถ๐˜๐—ถ๐—ป๐—ด Fix a matra, swap a conjunct or correct a single character in fewer taps. ๐ŸŽจ๐—ฃ๐—ฒ๐—ฟ๐˜€๐—ผ๐—ป๐—ฎ๐—น๐—ถ๐˜‡๐—ฒ๐—ฑ ๐—ธ๐—ฒ๐˜†๐—ฏ๐—ผ๐—ฎ๐—ฟ๐—ฑ ๐—น๐—ฎ๐˜†๐—ผ๐˜‚๐˜๐˜€ Arrange the keyboard the way you like to type. โšก๐—™๐—ฎ๐˜€๐˜๐—ฒ๐—ฟ ๐—ฒ๐˜ƒ๐—ฒ๐—ฟ๐˜†๐˜„๐—ต๐—ฒ๐—ฟ๐—ฒ Under the hood, we've made swipe, predictions and the whole app noticeably faster. โœ๏ธAnd of course, gesture swipe is still at the core, across multiple scripts, with Roman output for Hinglish, Tanglish and more. ๐Ÿ“ฒ๐——๐—ผ๐˜„๐—ป๐—น๐—ผ๐—ฎ๐—ฑ ๐—ผ๐—ป ๐—”๐—ป๐—ฑ๐—ฟ๐—ผ๐—ถ๐—ฑ: play.google.com/store/apps/dโ€ฆ ๐ŸŽ๐—ถ๐—ข๐—ฆ ๐˜‚๐˜€๐—ฒ๐—ฟ๐˜€, ๐˜†๐—ผ๐˜‚'๐—ฟ๐—ฒ ๐—ป๐—ฒ๐˜…๐˜! IndicSwipe for iPhone is coming in a few weeks. Stay tuned. This is still early, so tell us what you think in the comments or in a Play Store review. ๐—ฌ๐—ผ๐˜‚๐—ฟ ๐—ณ๐—ฒ๐—ฒ๐—ฑ๐—ฏ๐—ฎ๐—ฐ๐—ธ ๐˜€๐—ต๐—ฎ๐—ฝ๐—ฒ๐˜€ ๐˜„๐—ต๐—ฎ๐˜ ๐˜„๐—ฒ ๐—ฏ๐˜‚๐—ถ๐—น๐—ฑ ๐—ป๐—ฒ๐˜…๐˜. ๐Ÿ™ ๐—•๐˜‚๐—ถ๐—น๐˜ ๐˜„๐—ถ๐˜๐—ต โค๏ธ ๐—ฏ๐˜† @srihari_118 @yamnaht @akkiisfrommars @n_narasim456 @prajdabre @MiteshKhapra @anoopk @manickam24 Saranya Sukumaran @ai4bharat @WSAI_IITM @iitmadras #IndicSwipe #AI4Bharat #IndicTrans2 #IndicConformer #IndicXlit #IndicLanguages #MadeInIndia
1
2
9
403
Thanmay retweeted
It's 2018 again. Everyone is excited about all these crazy images generated from GANs. The same guy that made GANs just came out with a deep learning textbook with Bengio. NIPS just changed their name to NeurIPS and you wonder if that's actually going to stick. You train your Encoder-Decoder LSTM model with attention. Some crazy people are starting to use this new model architecture called a Transformer. Other crazy people are starting to fine-tune with something called REINFORCE to maximize BLEU and ROUGE scores. OpenAI just solved a Rubik's cube with a robot hand. Life is good.
21
33
4
525
27,390
Thanmay retweeted
43 million people around the world are blind. India alone is home to about 8 million of them, and Braille literacy is estimated at only around 1%. The situation is similar across the globe. And it is not only them who struggle; their families do too. Every day, frontier AI models set new records in math, coding and science. But there is one thing they still can't do well: read Braille. "What is the point of reading Braille by sight?" I know some of you will ask this. Braille is a tactile language, made to be understood by touch. But this work helps the people around a Braille reader: family members, friends, teachers and educators can now understand what the child is reading. It also helps Braille transcribers, and lets teachers guide their students better. Today we are introducing our work, "Learning to Read Braille: Benchmarking and Improving Braille Recognition Capabilities of VLMs." We built BrailleBench, a benchmark of 259 real-world Grade-1 and Grade-2 Braille images. We also designed a synthetic Braille generation pipeline with diverse layouts and image perturbations, and trained BrailleQwen, a 2B VLM, on this synthetic data. BrailleQwen reaches 87% exact match on Grade-1 and 61% on Grade-2 BrailleBench, outperforming standalone VLMs. We also tested it on a private Braille book dataset, where it achieved a CER of 0.013, a WER of 0.04, and an exact match of 94.37%. While the vision aspect of Braille still needs improving, the next challenge will be combining it with speech for true inclusivity for people with vision impairment. More to come! ๐Ÿ“„ arxiv.org/abs/2610.36195 ๐Ÿ’ป github.com/kartik0806/Braillโ€ฆ ๐Ÿค— huggingface.co/ai4bharat/Braโ€ฆ This work was led by Atharva Gawane and Kartik Jangra in @ai4bharat with guidance from Prof Hema Karah of IIT Madras who is visually impaired himself and has given us invaluable insights.
8
19
1
80
3,808
Interviewerย : toh seedha seedha nahi keh sakte they? Thodi der pehle koshish ki thi Sir, lekin aapko seedha seedha pasand nahi aaya.
q: why can't we just remove positional embeddings from transformers? bad answer: "because transformers need to know word order" good answer: "self-attention without positional information is permutation equivariant. rearranging the input tokens rearranges the outputs, but doesn't tell the model which token came first" understand the math, knowing the architecture is not enough :)
1
96
Hi My grandmother is admitted in Holy Family Hospital (share.google/dRiocgvTuR75K7Mโ€ฆ) and is in need of AB- blood. If you or anyone you know is around and open to donating that would be very very helpful.
2
8
11
518
Replying to @0x_Shashi
If God had given you some skills to read, you would compare both of those repositories and see what he has built. He has built an LLM fine-tuned on SMS messages. I have built a regex parser. If you could see that expense trackers have been there for many years, that would take some critical thinking to actually compare and check, but no, let's just see and just blame people without thinking twice. Additionally, learn some Hindi. The word "expense" translates literally to kharcha. Your friend has not minted the word kharcha, and it's used publicly across many expense trackers. Please go and disturb them instead.
2
2
4
46
8,016
If youre going to fail then fail fast but donโ€™t fail trying.
If youre going to fail then fail fast.
5
187
(1/2) Can we simulate garments without a fixed mesh or global 2D parameterisation? Introducing NeuralGarSim ๐Ÿ‘•: our #ECCV2026 work on neural garment simulation directly in 3D using Tangential Differential Calculus. With @NavamiK, @VGolyanik & C. Theobalt๐Ÿ‘‡
1
2
10
1,098
๐Ÿ“ฃ The ICASSP 2027 Speech and Language Processing reviewer nomination form is now open! Qualified researchers are invited to self-nominate and help shape the ICASSP 2027 technical program. Review the eligibility requirements and apply here: lnkd.in/etH2wKSv #ICASSP2027
13
1
14
3,754
Thanmay retweeted
helping GPT6 Astra become agi
430
2,122
266
53,294
1,577,350
Everything we have built at @Bodhan_AI and @ai4bharat is now live. ๐ŸŽ‰ Open-weight speech, vision & translation models for India's languages, available on Hugging Face. The same stack, hosted at some of the lowest prices in the market. ๐Ÿค— huggingface.co/bodhan-ai Read more about in our blogs: >>> bodhan.ai/research/blogs API end-points live now on Bodhan AI. >>> console.bodhan.ai/ Shout-out to @SarvamAI, our inference infrastructure partners for helping us serve our models reliably and affordably at scale. Dedicated to Pushpak Bhattacharyya Sir. His love for education and Indian language technology inspired all of it. ๐Ÿ™
27
77
8
488
133,574
This year was special: we had a submission in each EMNLP track: 1 conference, 1 tutorial, 1 demo. All three didn't make it, for reasons that were... interestingly unrelated to the content itself. Didnโ€™t go our way, but glad we got to do three very different kinds of things :)
1
5
948
Thanmay retweeted
Important for anyone who did not get into EMNLP. You can still commit to AACL ๐Ÿ”ฝ
๐Ÿ“ฃ The second round of commitment for IJCNLP-AACL 2026 will be open from August 18 to August 25, 2026 via the commitment link below. Acceptance decisions will be announced together on September 7, 2026. #NLProc #AACL2026 #IJCNLP2026 @aadi_joshi @KTLIM openreview.net/group?id=aclwโ€ฆ
1
1
239
Need this confidence in life
See you at EMNLP! We have several papers there! Iโ€™ll be there for a short visit, so feel free to message me. #EMNLP
1
7
769
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?
I got the same kind of review. Not only does the reviewer discriminate based on experience but also based on institution.
2
9
1,674
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.
๐Ÿš€ Excited to release IndicSwipe! ๐Ÿ‡ฎ๐Ÿ‡ณ 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. ๐Ÿ“ฑ Play Store: play.google.com/store/apps/dโ€ฆ ๐Ÿ’ป GitHub: github.com/AI4Bharat/Indic-Sโ€ฆ More features coming soon! ๐Ÿ“ Share your feedback: forms.gle/BG8btXNj9LR5qZmeA @srihari_118 @yamnaht @prajdabre @MiteshKhapra @ai4bharat @WSAI_IITM @iitmadras #IndicSwipe #OpenSource #AndroidDevelopment #AndroidDev #MachineLearning #Kotlin #IndicLanguages #AI4Bharat #MobileAppDevelopment #TechForIndia #OpenInnovation
3
255
TIL Jakarta has the unfortunate title of the worldโ€™s fastest sinking city i thought metaphorically โ€œsinkingโ€ at first and then i googled it๐Ÿ’€
Replying to @yamnaht
Coz it's sinking?
1
168
TIL Indonesiaโ€™s capital is no longer Jakarta. hah kok gitu sih
1
1
144
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.
1
6
1,071