How divyajyoti019/nlp-cross-lingual-project works
IndicRAG-QA is a multilingual education-policy question-answering demo. Users submit questions through the browser app or CLI; the system retrieves and ranks policy passages, applies an answerability threshold, then returns an evidence-grounded answer or abstains, with citations. Document ingestion creates searchable passages, while evaluation measures retrieval, answer quality, and abstention. The graph also preserves documented browser capabilities and optional integrations. Some UI wiring and evaluation details are unsampled, so those relationships are shown only where the README or excerpts support them.
Main entry points: pipeline.py QA orchestration · app.py Browser app · cli.py CLI commands · retrieval.py Passage retrieval
Components 16
User interfaces
- Browser app
app.py - CLI commands
src/indicrag/cli.py - Method comparison
app.py - Corpus explorer
app.py - Document upload
app.py
Question answering
- QA orchestration
src/indicrag/pipeline.py - Passage retrieval
src/indicrag/retrieval.py - Answer generation
src/indicrag/generation.py - Passage schema
src/indicrag/schema.py
Corpus and ingestion
- Document ingestion
src/indicrag/ingest.py - Starter passages
data/sample/passages.jsonl - Expanded passages
Evaluation
- Evaluation metrics
src/indicrag/evaluation.py - QA evaluation set
data/eval/qa.jsonl
Other components
- Questioner
- Ollama service
Connections 21
- Questioner → Browser app: asks via
README.md - Questioner → CLI commands: asks via
README.md - Browser app → QA orchestration: submits questions
README.md - CLI commands → QA orchestration: runs ask
src/indicrag/cli.py - QA orchestration → Passage retrieval: searches passages
src/indicrag/pipeline.py - QA orchestration → Answer generation: generates answer
src/indicrag/pipeline.py - QA orchestration → Passage schema: uses passage data
src/indicrag/pipeline.py - Passage retrieval → Passage schema: ranks passages
src/indicrag/retrieval.py - Answer generation → Ollama service: calls when enabled
src/indicrag/generation.py - Browser app → Method comparison: offers comparison
README.md - Method comparison → Passage retrieval: compares methods
README.md - Browser app → Corpus explorer: offers search
README.md - Browser app → Document upload: offers uploads
README.md - Document upload → Document ingestion: ingests documents
README.md - CLI commands → Document ingestion: runs ingestion
src/indicrag/cli.py - Document ingestion → Passage schema: creates passages
src/indicrag/ingest.py - Starter passages → QA orchestration: loads corpus
README.md - Expanded passages → QA orchestration: loads optionally
README.md - CLI commands → Evaluation metrics: runs evaluation
src/indicrag/cli.py - Evaluation metrics → Passage retrieval: scores retrieval
src/indicrag/evaluation.py - Evaluation metrics → QA evaluation set: reads questions
README.md