How abhijitmanepatil/indiannlp-transliteration works
IndianNLP-Transliteration is a neural transliteration project with a deployable Python/API interface and research workflows for training and evaluating language models. The principal user flow is text and language input through the HTTP API into the configured transliteration engine, which returns candidate spellings; the API also accepts user-choice and annotation feedback. The README additionally highlights model training and language datasets. Source excerpts are partial, so links to unsampled model-serving and training subsystems are limited to relationships documented by the README or shown in sampled code.
Main entry points: api_expose.py HTTP API · tasks Training runners · infer_engine.py Inference engine · retrieve_wikidata_xlits.py Wikidata pairs
Components 16
Transliteration serving
- HTTP API
apps/api_expose.py - Library server
apps/ai4bharat/transliteration/xlit_server.py - Transliteration engine
apps/ai4bharat/transliteration/xlit_src.py - Language lineup
apps/ai4bharat/transliteration/models/default_lineup.json - Feedback logs
apps/api_expose.py - Cloud feedback store
Model research
- Training runners
tasks - Inference engine
tasks/infer_engine.py - Neural architectures
algorithms - Language model fusion
algorithms/lm_fused_rnn.py - Training utilities
utilities
Language data
- Transliteration datasets
data - Dataset conversion
tools/json_handler.py - Wikidata pairs
tools/wiki/retrieve_wikidata_xlits.py
Other components
- API user
- Wikidata service
Connections 14
- API user → HTTP API: sends text
- HTTP API → Library server: gets app
- Library server → Transliteration engine: serves transliteration
- Transliteration engine → Language lineup: loads configuration
- HTTP API → Feedback logs: writes feedback
- HTTP API → Cloud feedback store: optionally stores
- Training runners → Neural architectures: uses architectures
- Training runners → Language model fusion: uses fusion model
- Training runners → Training utilities: uses utilities
- Inference engine → Neural architectures: runs models
- Inference engine → Training utilities: uses utilities
- Dataset conversion → Transliteration datasets: converts data
- Wikidata pairs → Wikidata service: queries
- Wikidata pairs → Transliteration datasets: writes pairs