How etienne-landryb/docviz-document-intelligence-tool works
DocViz is a Streamlit financial-document intelligence tool. Its main workflow takes an uploaded digital PDF through coordinate-aware parsing, metric extraction and normalization, validation, and charts that represent missing figures as undisclosed. A separate ingestion and Q&A workflow embeds prose and metric facts, stores them for similarity retrieval, and abstains when retrieved context is weak. Comparison templates support FXAIX/VFIAX demos, while the vector-store layer provides document management. The README and excerpts support these capabilities; some app orchestration details and unsampled modules are documented rather than directly shown.
Main entry points: app.py Streamlit app · ingest.py Chunk ingestion · rag.py Grounded Q&A · store.py Vector store
Components 18
User workflow
- Streamlit app
app.py
Metric intelligence
- PDF parser
docviz/parse.py - Metric extraction
docviz/extract.py - Number normalization
docviz/normalize.py - Metric contract
docviz/schema.py - Sanity gate
docviz/validate.py - Chart arrangement
docviz/arrange.py - Chart renderer
docviz/viz.py
Retrieval and Q&A
- Chunk ingestion
docviz/ingest.py - Local embeddings
docviz/embed.py - Vector store
docviz/store.py - Grounded Q&A
docviz/rag.py
Comparison
- Fund templates
docviz/funds.py - Fund comparison
run_phase1_compare.py
Other components
- User
- Digital PDF
- Groq API
- Supabase pgvector
Connections 24
- User → Streamlit app: uploads and asks
README.md - Streamlit app → PDF parser: parses PDF
app.py - Digital PDF → PDF parser: provides pages
docviz/parse.py - Streamlit app → Metric extraction: extracts metrics
README.md - Metric extraction → Groq API: requests extraction
README.md - Metric extraction → Number normalization: parses values
- Streamlit app → Number normalization: normalizes values
app.py - Streamlit app → Sanity gate: validates metrics
README.md - Sanity gate → Metric contract: checks metrics
docviz/validate.py - Streamlit app → Chart renderer: requests charts
app.py - Chart renderer → Chart arrangement: arranges panels
docviz/viz.py - Streamlit app → Chunk ingestion: ingests document
- Chunk ingestion → PDF parser: reads PDF rows
docviz/ingest.py - Chunk ingestion → Fund templates: extracts metric facts
docviz/ingest.py - Chunk ingestion → Local embeddings: embeds chunks
docviz/ingest.py - Chunk ingestion → Vector store: writes chunks
docviz/ingest.py - User → Grounded Q&A: asks question
README.md - Grounded Q&A → Local embeddings: embeds question
docviz/rag.py - Grounded Q&A → Vector store: searches chunks
docviz/rag.py - Vector store → Supabase pgvector: reads and writes
docviz/store.py - Grounded Q&A → Groq API: grounds answer
docviz/rag.py - Fund comparison → Fund templates: extracts funds
run_phase1_compare.py - Fund comparison → Chart renderer: renders comparison
run_phase1_compare.py - User → Vector store: manages documents