Bibliography QA Assistant

Inspiration

Research bibliography and yearbook spreadsheets are assembled over years by multiple people. Before the data can be trusted, researchers need to find missing abstracts, inconsistent review statuses, and records that may refer to the same work—without risking the original file.

What it does

Bibliography QA Assistant is a local-only, read-only spreadsheet quality-review MVP. It recognizes common title, abstract, status, author, and year columns, then creates a risk-sorted human review queue. It flags blank abstracts, abstract/status contradictions, and possible same-title records, and exports CSV or HTML review reports.

The public demo uses only a tracked synthetic workbook. The tool never uploads, changes, deletes, merges, overwrites, or auto-completes source records. Similarity results are always prompts for human confirmation, never automatic duplicate decisions.

How we built it

The project uses Python, openpyxl, JavaScript, HTML, and CSS. It runs locally with no account, API key, or cloud storage. A synthetic demo mode makes the public workflow reproducible without opening a researcher’s workbook.

Codex accelerated the local-only architecture, implementation, test coverage, release hygiene, and documentation. GPT-5.6 identified a status-classification edge case: Not complete / 未完成 could be misread as completed by substring matching. We corrected that rule and added regression tests while preserving the human-only duplicate decision boundary.

Challenges we ran into

The challenge was making the automation useful without letting it become an unsafe data-cleaning tool. A review queue had to expose risk clearly while protecting source workbooks from alteration.

Accomplishments that we're proud of

We created a working local MVP that turns a synthetic bibliography workbook into a transparent, risk-ranked review report in seconds, with no cloud upload and no automatic data changes.

What we learned

For scholarly data, transparent review support is more appropriate than automatic deletion or merging. The system can prioritize attention; the researcher retains the final judgment.

What's next for Bibliography QA Assistant

Next steps are broader column aliases, more configurable review rules, and additional local report formats—while keeping the same read-only, human-in-control safety boundary.

Try it

Clone the public repository, install the one dependency, and run python local_app.py --port 8765. The README includes a repeatable ?demo=1 route and the synthetic test workbook.

Built With

• css
• gpt-5.6
• html
• javascript
• openpyxl
• python