aakashMeghwar01/aakashmeghwar01

---

🧠 About Me

aakash = {
    "name"       : "Aakash Meghwar",
    "location"   : "Nizhny Novgorod, Russia 🇷🇺  ←→  Pakistan 🇵🇰",
    "education"  : "MSc @ HSE — Higher School of Economics (2024–)",
    "mission"    : "Bridge linguistics and AI for low-resource languages",
    "research"   : [
        "SindhiLM — Sindhi Language Model (Qwen2.5 fine-tune)",
        "SindhiLM-Tokenizer — Morpheme-aware BPE for Sindhi",
        "Sindhi Corpus 505M — largest open Sindhi pretraining corpus",
        "sindhinltk — first open-source Sindhi NLP Python library",
        "Sentiment Analysis · Computational Stylistics · CL",
    ],
    "languages"  : {
        "Sindhi" : "Native 🌙",
        "Urdu"   : "Native ✨",
        "English": "C2 🔥",
        "Russian": "A2 📖",
    },
    "looking_for": "PhD position in NLP / Computational Linguistics 🎓",
    "fun_fact"   : "Building NLP tools for 80M Sindhi speakers from scratch",
}

---

🚀 Flagship Projects

🤖 SindhiLM Sindhi Language Model — fine-tuning Qwen2.5-0.5B on 505M token Sindhi corpus. First transformer model dedicated to Sindhi.	🔤 SindhiLM-Tokenizer Morpheme-boundary-aware BPE tokenizer merged into Qwen2.5. v1 live · v2 with SindhiNLTK pre-segmentation coming.
📦 sindhinltk First open-source Python NLP toolkit for Sindhi. Zero dependencies. Pure Python. Works immediately.	📊 Sindhi Corpus 505M Largest open-source deduplicated Sindhi pretraining corpus. 742K docs · ~505M tokens · 11 source datasets.

---

📦 sindhinltk — Quick Start

pip install sindhinltk

from sindhinltk.tokenizer import SindhiTokenizer
from sindhinltk.sentiment import SindhiSentiment
from sindhinltk.stopwords import SindhiStopwords
from sindhinltk.stemmer   import SindhiStemmer

text = "سنڌي ٻولي تمام سٺي ۽ قديم آهي"

tokens = SindhiTokenizer().tokenize(text)
# → ['سنڌي', 'ٻولي', 'تمام', 'سٺي', '۽', 'قديم', 'آهي']

label  = SindhiSentiment().analyze(text)
# → 'مثبت'  ✅ positive

clean  = SindhiStopwords().remove_stopwords(tokens)
# → ['سنڌي', 'ٻولي', 'سٺي', 'قديم']

stems  = SindhiStemmer().stem_tokens(clean)
# → ['سنڌي', 'ٻولي', 'سٺي', 'قديم']

---

🗂️ All Repositories

🤖 Models

Model	Description	Stats
SindhiLM-Qwen-0.5B	Sindhi LM base · Text Generation
SindhiLM-Qwen-0.5B-SFT	SFT fine-tune	🔜
SindhiLM-Qwen-0.5B-V2	V2 improved
urdu-poetry-transformer	Urdu poetry generation	✅
aurat-march-sentiment-minilm	Sentiment · MiniLM

🔤 Tokenizers

Tokenizer	Description	Status
SindhiLM-Tokenizer-v1	BPE · +7,978 Sindhi tokens · merged into Qwen2.5	✅ Live
SindhiLM-Tokenizer-v2	Morpheme-aware · +4,571 cleaner tokens · SindhiNLTK pre-seg	🔜 Soon

📊 Datasets

Dataset	Description	Size
sindhi-corpus-505m	Largest open Sindhi pretraining corpus	742K docs · 505M tokens
Sindhi-Intelligence-Core-SFT	SFT instruction dataset	361K rows
Sindhi-Intelligence-Core-SFT-V2	SFT V2 improved	46.1K rows
Sindhi-Factual-QA	Factual QA pairs	99 rows
sindhi-tokenized-505m	Pre-tokenized corpus	742K rows
aurat-march-processed-news	News corpus for discourse analysis	5K rows

---

📚 Publications

📄 [Springer, 2026] Compact Transformer Models for Classical Urdu Poetry: A Computational Stylistics Approach Language Resources and Evaluation — Transformers Urdu NLP Computational Stylistics

📄 [SSRN, 2025] The Comparative Analysis of How the Aurat March Movement is Represented in British and Pakistani Pro-Government Press Read on SSRN → — Critical Discourse Analysis Corpus Linguistics

🎤 [April 2025] Sentiment Analysis of International Students' Online Reviews 78th Herzen Readings, St. Petersburg, Russia — Sentiment Analysis NLP

---

🛠️ Tech Stack

---

🔭 Currently Working On

▸ 🏋️  Fine-tuning SindhiLM — Qwen2.5-0.5B on Sindhi Corpus 505M
▸ 🔤  SindhiLM-Tokenizer-v2 — morpheme-aware BPE + SindhiNLTK pre-segmentation
▸ 📦  sindhinltk v1.4 — NER + POS tagger modules
▸ 🎓  Seeking PhD position in NLP / Computational Linguistics
▸ 📝  Writing research on low-resource Sindhi language modeling

---

💡 Research Interests

Low-Resource NLP · Language Modeling · Morphological Analysis · BPE Tokenization · Arabic-Script NLP · Computational Stylistics · Multilingual Transformers · Sindhi & Urdu NLP · Fine-tuning LLMs · Corpus Linguistics · Sentiment Analysis

---

🤝 Open For

• 🎓 PhD collaborations in NLP / Computational Linguistics
• 🔬 Research partnerships on low-resource South Asian languages
• 🛠️ Contributions to sindhinltk and SindhiLM
• 💬 Discussions on Sindhi · Urdu · Arabic-script NLP

📧 aakashmeghwar01@gmail.com · 🌐 texttechsolutions.com

---

"There are 7,000 languages in the world. Most have zero NLP tooling. Let's fix that."

— Aakash Meghwar, building for 80 million Sindhi speakers