Silappatikaram · Manimekalai · Seevaga Sinthamani · Valayapathi · Kundalakesi · Ga (Muruga — தமிழின் தோற்றம்)
SiMaSeVaKuGa is a homegrown, sovereign Large Language Model (LLM) — a vLLM/sLLM platform delivering cutting-edge language model capabilities tailored specifically for the Tamil language and Indian linguistic contexts.
A robust model capable of understanding, generating, and reasoning in Tamil — built by Tamil people, for Tamil people.
Named after the Five Great Tamil Epics (ஐம்பெரும் காப்பியங்கள்) and Ga honouring Muruga — deeply intertwined with the origin of Tamil — this model is trained on the richest corpus of Tamil literary heritage ever assembled for machine learning.
Tamil is one of the world's oldest living classical languages. Yet Tamil speakers — 80+ million people — are underserved by English-centric global LLMs that fail to capture Tamil's grammatical depth, dialectal richness, and cultural nuance.
SiMaSeVaKuGa LLM changes that:
| Problem | SiMaSeVaKuGa Solution |
|---|---|
| Global LLMs hallucinate in Tamil | Trained natively on Tamil corpus |
| No Tamil-first GenAI tooling | Open API for Tamil GenAI & Agentic AI |
| English-centric reasoning | Tamil-optimized reasoning & generation |
| No Tamil voice AI | Dialect-aware voice inference |
| Tamil culture lost in translation | Trained on classical Tamil literature |
The model is trained on Tamil's documented literary and linguistic heritage:
| Source | Description | Est. Date |
|---|---|---|
| Tolkappiam (தொல்காப்பியம்) | World's oldest surviving Tamil grammar — 3 chapters: Ezhuthu, Soll, Porul | 500 BCE – 300 CE |
| Silappatikaram (சிலப்பதிகாரம்) | 1st of the Five Great Epics | ~2nd century CE |
| Manimekalai (மணிமேகலை) | 2nd of the Five Great Epics | ~5th century CE |
| Seevaga Sinthamani (சீவகசிந்தாமணி) | 3rd of the Five Great Epics | ~10th century CE |
| Valayapathi (வளையாபதி) | 4th of the Five Great Epics | ~10th century CE |
| Kundalakesi (குண்டலகேசி) | 5th of the Five Great Epics | ~5th century CE |
| Tirukkural (திருக்குறள்) | 1,330 couplets — ethics, governance, love | ~300 BCE – 500 CE |
| Sangam Literature | Akananuru, Purananuru, Natrinai, and more | 300 BCE – 300 CE |
| Contemporary Tamil | News, social media, government documents, spoken dialects | Present |
┌─────────────────────────────────────────────────────────┐
│ SiMaSeVaKuGa LLM Capabilities │
├─────────────────────────────────────────────────────────┤
│ 🧠 Tamil Language Understanding & Generation │
│ 💬 Conversational AI (chatbots, assistants) │
│ 🤖 Agentic AI — Tamil-native AI agents │
│ 🎙️ Voice-first, dialect-aware inference │
│ 📄 Document summarisation & translation │
│ ⚖️ Legal & governance document processing │
│ 🏥 Healthcare advisory in Tamil │
│ 🌾 Agricultural advisory (rural dialects) │
│ 🎓 Education & e-learning platforms │
│ 🏛️ E-governance & citizen services │
└─────────────────────────────────────────────────────────┘
SiMaSeVaKuGa LLM
├── Base Model → Transformer-based, Tamil-first pre-training
├── vLLM Layer → High-throughput inference (PagedAttention)
├── sLLM Variants → Lightweight models for edge/mobile deployment
├── Tamil Tokenizer → Custom BPE tokenizer for Tamil script
├── Voice Module → Dialect-aware ASR + TTS
└── API Gateway → REST + gRPC inference endpoints
| Phase | Timeline | Milestone |
|---|---|---|
| Phase 1 — Foundation | Months 1–12 | Corpus assembly, tokenizer, base pre-training |
| Phase 2 — Acceleration | Months 12–24 | Fine-tuning, RLHF, public API beta |
| Phase 3 — Scale | Months 24–36 | Production deployment, edge models, voice API |
Developers and enterprises can use SiMaSeVaKuGa LLM to build:
- GenAI applications — Tamil content generation, summarisation, Q&A
- Agentic AI systems — autonomous Tamil-language agents for workflows
- Government services — citizen-facing Tamil AI for e-governance
- Healthcare — Tamil diagnostic assistants and patient communication
- Education — Tamil tutoring systems and adaptive learning
- Agriculture — voice-based advisory for Tamil farmers
# Clone the repository
git clone https://github.com/tvkvolunteers/simasevakuga-llm.git
cd simasevakuga-llm
# Install dependencies
pip install -r requirements.txt
# Run inference (once model weights are available)
python inference.py --prompt "தமிழ்நாட்டின் தலைநகரம் எது?"from simasevakuga import TamilLLM
llm = TamilLLM(model="simasevakuga-7b")
response = llm.generate(
prompt="தமிழ்நாட்டின் வரலாற்றை சுருக்கமாக கூறு",
max_tokens=512,
dialect="standard" # or "madurai", "coimbatore", "jaffna"
)
print(response.text)We welcome contributions from Tamil developers, linguists, researchers, and AI engineers worldwide.
1. Fork the repository
2. Create your feature branch: git checkout -b feature/your-feature
3. Commit your changes: git commit -m 'Add Tamil corpus segment'
4. Push to the branch: git push origin feature/your-feature
5. Open a Pull Request
Areas where we need help:
- Tamil corpus curation and cleaning
- Tokenizer improvements for Tamil script
- Dialect data collection (Madurai, Coimbatore, Jaffna, Eelam Tamil)
- Evaluation benchmarks for Tamil NLP
- Fine-tuning datasets for specific domains
Apache License 2.0 — open for Tamil people and the world.
Built with pride by TVK Volunteers — an independent volunteer network supporting Thamizhaga Vettri Kazhagam and the people of Tamil Nadu.
உழைப்பு மக்களுக்கு • வெற்றி TVK-க்கு For the People. For TVK.
- 🌐 Website: tvkvolunteers.org
- 🐦 X / Twitter: @TVKvolunteers
- 📺 YouTube: @TVKvolunteers
- 💼 LinkedIn: tvk-volunteers