Projects / prototype

AI assistance for understanding university lab reports, checking source evidence and reviewing rubric-linked suggestions. Final scores and feedback stay with the instructor.

Updated2026-10-26

Complete source ZIP, not a standalone installer. Follow the repository setup guide for Python dependencies, model configuration and optional Word conversion. Windows setup still needs separate validation. No student records, databases or secrets are included.

Problem

Reviewing an English lab report means understanding scientific reasoning, checking assignment requirements, locating evidence and writing useful feedback across several tools. Labwise brings those steps into one teacher workspace.

Architecture

Assignment + rubric snapshot → DOCX parsing and source anchors → background AI candidates → linked source, outline and teacher review → instructor confirmation → Word feedback package.

Design philosophy

Understand before assessing. Every candidate should point back to a real paragraph, quotation or figure object and the current rubric. AI completion and teacher completion are separate states.

Workflow

1. Create an assignment and verify the rubric
2. Import DOCX reports and queue analysis
3. Read the source alongside evidence-linked candidates
4. Accept, edit or reject; enter scores
5. Confirm feedback and continue reviewing
6. Export the instructor-approved review package

Implementation

The local pilot includes assignment/rubric maintenance, DOCX import, automatic analysis queues, evidence-linked suggestions, draft/revision persistence, continuous review and instructor-confirmed Word review packages. New student submissions are DOCX-only; historical PDF records remain readable. This remains a local instructor pilot. Screenshots use fictional test material and are not acceptance evidence for every current feature.

Reflection & next steps

The quality goal is faster understanding and better-grounded feedback, not automated final grading. Evaluate evidence accuracy, rubric alignment, unsupported suggestions and reasons for teacher edits before expanding the pilot. The current snapshot has known regression failures, is not production-ready, and makes no validated percentage claim of efficiency gains.

Technology

PythonJavaScriptSQLiteDOCX / Word commentsDeepSeekPDF.js

Interface specimens