MD Mizanur Rahman

Work · Retrieval and graphs

Graph provenance explorer: source-preserving retrieval

  • Demo ready
  • Synthetic demonstration
  • Companion to coauthored research

A companion explorer that builds a typed graph over a small fictional corpus, retrieves passages through lexical and graph-neighbourhood paths, and renders answers that cite only identifiers that resolve.

Open the interactive demonstrationRepository on GitHub

Problem

A retrieval-augmented summary is only as trustworthy as the path from each sentence back to a source. The explorer makes that path visible: which passage, through which typed relation, with what provenance.

Contribution

Relationship to research: Companion to coauthored research. Newly built companion explorer for a coauthored article on knowledge-graph retrieval-augmented clinical decision-support summaries. It is not the article's original code, and retrieval quality on this synthetic corpus says nothing about the article's system. The process is described on its own terms; no author is named here.

Account owner’s role: Commissioned, reviewed and published the implementation. The code was produced with AI-assisted generation on commission and reviewed before release; it is not the original code of any cited publication.

What was built

  • Ingestion of twelve authored fictional guidance notes into typed document, passage, concept, assertion and source nodes
  • In-repository BM25 baseline and one-hop typed graph expansion with reranking
  • Deterministic answer renderer with abstention when no passage clears the threshold
  • Fixed-seed layout export so the web explorer shows the identical arrangement

Method

Seed passages come from lexical scoring; the graph step expands one hop along typed edges and reranks by lexical score plus an edge-type weight. Every returned passage carries its path.

Data and access

Authored fictional corpus with invented index names and workflow terms. No guidelines, no patient reports, no publisher text.

Evaluation protocol

recall@k, MRR, source-path validity, unsupported-citation rate and abstention accuracy on twenty answerable and five unanswerable fixture questions.

Results

No measured result is published for this project. Demonstration outputs from the repository’s own synthetic fixtures are labelled as such in the repository and are not presented here as results.

Limitations

  • A visual explorer does not establish that the published system's retrieval was quantitatively validated.
  • The corpus is tiny and fictional by design.

Reproducibility

Offline; graph export and metrics regenerate from the corpus with a fixed seed.

Repository metadata: Python, MIT, last push 2026-09-13, 0 stars.

Source publication and links