Skip to content

Contributing

Contributions are welcome. Here is what is most needed.

What to contribute

1. Legacy font additions

If you have a Nepali PDF with a font LamiSema does not detect, open an issue with the font name. You can find the font name by running:

pdffonts your-file.pdf

Or with Python:

import fitz
doc = fitz.open("your-file.pdf")
for page in doc:
    for font in page.get_fonts(full=True):
        print(font[3])  # base font name

Include a sample PDF (with PII removed) if possible.

Additional regex patterns for Nepali entity types. The v1.2 roadmap tracks these priority types — contributions that implement and test any of them are especially welcome.

Note: Since v1.0, LamiSema is language-agnostic. You can also contribute backends for other scripts (e.g. Hindi, English).

Priority Nepali types: - WARD_CODE — वडा नं. X (ward number) - VDC_MUNICIPALITY — गाउँपालिका / नगरपालिका names (753 local units) - DISTRICT — all 77 Nepali districts ... - PROVINCE — 7 province names - LAND_PARCEL — कि.नं. (kittaa number), land categories (ऐलानी, रैकर, गुठी) - PERSON_NAME_NE — Nepali personal names - PHONE_NE — Nepali phone formats - COURT_CASE — मिसिल नं., मुद्दा नं. references - GAZETTE_REF — नेपाल राजपत्र volume and notice numbers - GOV_POSITION — government title patterns (सचिव, महानिर्देशक, etc.) - Vehicle registration numbers - National ID numbers (राष्ट्रिय परिचयपत्र)

Patterns belong in lamisema/nlp/ner.py. Each new entity type needs a corresponding test in tests/test_ner.py.

3. Benchmark PDFs

Anonymized Nepali PDFs of each encoding type for the evaluation dataset. See benchmark/dataset/README.md for sourcing guidelines. Contact the maintainer before submitting to coordinate dataset logistics.

4. BS→AD accuracy

The current Bikram Sambat → AD conversion is approximate (±1 day). Corrections and improvements to the conversion logic in lamisema/nlp/dates.py are welcome, especially:

  • Edge cases around Nepali New Year (mid-April)
  • Correct handling of BS months 10–12 (Magh–Chaitra) year rollover
  • Integration of the nepali-datetime library for exact conversion

Development setup

git clone https://github.com/sanjiblamichhane/lamisema
cd lamisema
pip install -e ".[dev]"

# Install Tesseract
brew install tesseract tesseract-lang   # macOS

# Run tests
pytest tests/ -v

# Run linter
ruff check lamisema/ tests/

Code standards

  • All new code must have docstrings
  • All new modules must have at least one test
  • New entity types need a test in tests/test_ner.py
  • New font names must be documented in docs/encoding-types.md
  • Run ruff check before opening a PR — CI will fail otherwise

Pull request checklist

  • [ ] Tests pass (pytest tests/ -v)
  • [ ] Linter passes (ruff check lamisema/ tests/)
  • [ ] New functionality has tests
  • [ ] CHANGELOG.md updated under [Unreleased]
  • [ ] Documentation updated if behaviour changed