Organize
Batch document sets are sorted into usable research collections before deeper processing begins.
BDS Platform
Batch Document Search
A local, private document-intelligence system that organizes complex files, indexes them for hybrid search, extracts entities, and builds an Obsidian-compatible knowledge graph.
The Pipeline
Four phases, from raw files to a knowledge graph.
BDS moves a document set through four stages — and each one is usable before the next completes.
Index
After Phase 2, hybrid BM25 and semantic search become available for immediate review.
Extract Entities
Phase 3a identifies people, organizations, places, and key terms for structured research review.
Build Knowledge Graphs
Phase 3b expands entity cards, relationships, and Obsidian-compatible graph exploration.
How To Read Results
Search signals, not magic scores.
A quick guide to the relevance signals BDS shows beside every result.
| Signal | Plain-English meaning |
|---|---|
| both | Meaning-match and word-match agree. This is the strongest relevance signal. |
| vector | Meaning-match. The text reads like the question even when different words are used. |
| bm25 | Word-match. It shares query words but may not actually answer the question. |
| CORPUS_GAP | The documents may not contain strong evidence for the question. |
Watch It Run · Ingestion
A real case, ingested end to end.
A 30-document random sample from the SEC v. Ripple docket moves through the pipeline — organize and OCR, chunk and embed, then entity extraction — captured locally at real speed. The same run produces the knowledge graph below.
Local capture: a 30-document SEC v. Ripple sample ingested end to end — organize/OCR, indexing, and entity extraction.
Hybrid Search · Synthesis
Ask the corpus. Read the signals.
Once a sample is indexed it's searchable. This capture runs hybrid semantic + keyword queries across two projects — the SEC v. Ripple and Enron corpora — showing each result row with the relevance signal behind it, then synthesizes an answer drawn only from the retrieved passages.
Local capture: hybrid semantic + keyword search across the SEC v. Ripple and Enron projects, result rows with relevance signals, and grounded answer synthesis.
Phase 3b · Knowledge Graph
The extracted entities, connected.
Once entities are pulled from the corpus, BDS writes them out as an Obsidian-compatible vault. This capture walks the same 30-document SEC v. Ripple sample: each node is a person, organization, or event, and each line is a relationship found across the filings — the same evidence, now navigable.
Local capture: the SEC v. Ripple entity graph in Obsidian — panning the full graph, then isolating a single entity to follow its links.