The IndiaML Tracker systematically identifies, analyzes, and highlights India's contributions to global machine‑learning research. Born from a Twitter exchange between Paras Chopra (Lossfunk Founder) and Sohan Basak (hard‑core technologist, building the future of human–AI interaction) in January 2025, the project focuses on research conducted within Indian institutions. By publishing transparent metrics, we aim to showcase domestic innovation and inspire the next generation of researchers.
Despite India's growing presence in top‑tier ML venues, there was no dedicated platform quantifying that contribution. IndiaML Tracker addresses this gap by providing institution‑level analytics grounded in openly verifiable data.
Our goals include:
- 🔍 Increasing visibility of Indian research institutions globally
- 📊 Creating benchmarks for measuring progress in ML research output
- 🤝 Identifying collaboration opportunities between institutions
- ✨ Inspiring young researchers by showcasing successful Indian contributions
Where does our data come from?
Currently, all paper metadata is sourced from the public OpenReview API. OpenReview is used by many—but not all—ML conferences. We therefore treat it as a significant subset, not the entirety, of relevant literature.
- Inclusion criterion: A paper is counted as “Indian research” when ≥ 1 author is affiliated with an Indian organisation at the time of publication.
- Optional filters:
- First‑author Indian: first listed author is India‑affiliated.
- Majority Indian: > 50 % of authors are India‑affiliated.
- Affiliation resolution: We use deterministic rules first (institution lookup tables), followed by LLM‑assisted disambiguation (via OpenRouter). Because LLMs are stochastic, errors can creep in—please help us correct them!
- Limitations:
- Conferences that do not publish to OpenReview are currently absent.
- Pre‑prints (e.g., arXiv) are outside our present scope.
- LLM hallucinations or missing metadata can introduce noise.
Found a missing or mis‑classified paper? Create an issue or open a pull request—we’ll review quickly.
IndiaML Tracker follows a modular, pipeline‑based architecture that enables systematic processing of research‑paper data:
Key design patterns:
- Adapter Pattern – standardises data collection from heterogeneous sources
- Factory Pattern – instantiates the correct adapter from config
- Repository Pattern – abstracts database operations
- Pipeline Pattern – chains discrete processing stages
Ongoing work: We are actively experimenting with additional data sources (e.g. ACL Anthology, arXiv bulk metadata) and more robust pipelines (e.g. deterministic disambiguation, structured affiliation ontologies) to keep improving coverage and accuracy.
For complete technical details, see Documentation.
| Purpose | Stack |
|---|---|
| Core language | Python 3.12+ |
| Storage | SQLite + SQLAlchemy ORM |
| Data source | OpenReview API |
| Affiliation resolution | LLM integration via OpenRouter |
We welcome contributions! The fastest way to help is to run the pipeline and submit data—this expands the dataset and validates existing entries.
# Clone the repository
git clone https://github.com/lossfunk/indiaml-tracker.git
cd indiaml-tracker
# (Recommended) set up with uv
uv venv --python=3.12
uv pip install .
# Alternative: standard venv
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt
# Add your API keys (for LLM steps)
echo "OPENROUTER_API_KEY=your_key_here" >> .env
# Run the pipeline step‑by‑step
python -m indiaml.pipeline.process_venue
python -m indiaml.pipeline.process_authors
python -m indiaml.pipeline.process_paper_author_mapping
python -m indiaml.pipeline.patch_unk_cc2
python -m indiaml.pipeline.patch_unk_cc3 # <-- inspect logs for unmatched affiliations
python -m indiaml.pipeline.patch_unk_cc4
# Optional LLM‑based PDF workflow
python -m indiaml.pipeline.patch_unk_cc5
# Analytics & output
python -m indiaml.analytics.analytics
python -m indiaml.pipeline.generate_final_jsons
python -m indiaml.pipeline.generate_summariesThen submit a pull request with the updated JSON files and summaries.
Data quality is paramount. You can help by reviewing:
- Author names / affiliations
- Institutional assignments
- Country codes
Submit corrections via PR or by opening an issue.
- 💡 Enhance affiliation‑resolution algorithms
- ➕ Add new data sources (arXiv, ACL Anthology…)
- 📚 Improve documentation
- 🐛 Fix bugs
See CONTRIBUTING.md for full guidelines.
- 📚 A continuously updated database of ML papers with Indian author affiliations
- 🏢 Institution‑level insights and publication trends
- 📈 Year‑over‑year tracking of India’s contribution to global ML research
- 🌐 Visualisation of collaborations between Indian and international institutions
- Python 3.8+
- SQLite
- Git
# 1. Clone
git clone https://github.com/lossfunk/indiaml-tracker.git
cd indiaml-tracker
# 2. Virtual env & deps
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt
# 3. Environment variables
echo "OPENROUTER_API_KEY=your_api_key_here" >> .env
# 4. Run tests
python -m unittest discover indiaml.testsIf you hit issues:
- Check the Troubleshooting section in DOCUMENTATION.md.
- Search existing GitHub issues.
- Open a new issue with details if it’s novel.
Code is released under the MIT License (see LICENSE).
Data notice: Some metadata originates from third‑party conference proceedings. While we are evaluating an open data licence compatible with those sources, the data itself may ultimately be published under a licence different from MIT to comply with all relevant laws and terms. We will document any change clearly.
Made with ❤️ by the IndiaML Tracker team — join us in highlighting India’s contributions to global ML research!