Source: https://github.com/aipoch/medical-research-skills

When to Use

• Evaluate a research paper, thesis, or proposal and produce a structured critique with scores.
• Generate actionable revision recommendations across core academic writing dimensions.
• Compare multiple drafts/versions of a manuscript using consistent rubric-based scoring.
• Assess submission readiness (e.g., for a conference/journal) and identify major weaknesses.
• Review a document provided as a PDF/DOCX/TXT file when the user expects automatic text extraction.

Key Features

• Automatic text extraction from PDF/DOCX/TXT via scripts/extract_text.py (intended as the first step for file inputs).
• ScholarEval rubric with 8 evaluation dimensions (see references/evaluation_framework.md).
• Per-dimension scoring (1–5) with qualitative feedback and concrete recommendations.
• Weighted score calculation via scripts/calculate_scores.py from a JSON score file.
• Produces a final report summarizing strengths, weaknesses, and next steps.

Dependencies

• Python 3.10+
• See requirements.txt for pinned Python package versions (install via pip install -r requirements.txt).

Example Usage

A) Evaluate a PDF/DOCX/TXT file (end-to-end)

1. Extract text (run this first for file inputs):

python scripts/extract_text.py "paper.pdf"

2. Create a scores JSON (example: scores.json):

{
  "problem_formulation": 4,
  "literature_review": 3,
  "methodology": 4,
  "data_quality": 3,
  "analysis": 4,
  "results": 3,
  "writing_quality": 4,
  "citations": 3
}

3. Compute the weighted/aggregate score:

python scripts/calculate_scores.py --scores scores.json

4. Use the extracted text plus the rubric to generate the evaluation report:

• Apply the 8-dimension criteria from references/evaluation_framework.md
• Provide per-dimension justification, then summarize strengths/risks and prioritized revisions

B) Evaluate pasted text (no extraction)

If the user pastes text directly (e.g., abstract, full paper text), skip extraction and evaluate immediately using the 8 dimensions and the 1–5 scale.

Implementation Details

File ingestion protocol (for PDF/DOCX/TXT)

• For any user-provided file, run:

  python scripts/extract_text.py "<filename-or-path>"

• The extraction script is designed to locate the file even if the full path is not provided.
• Use the extracted plain text as the sole input to the evaluation rubric and scoring.

Evaluation dimensions (8)

The framework evaluates:

1. Problem Formulation
2. Literature Review
3. Methodology
4. Data Quality
5. Analysis
6. Results
7. Writing Quality
8. Citations

Detailed criteria and guidance are defined in:

• references/evaluation_framework.md

Scoring scale (1–5)

• 1 — Poor: Major flaws; not usable as-is.
• 2 — Weak: Significant issues; major revision required.
• 3 — Average: Acceptable baseline; improvement needed.
• 4 — Good: Strong overall; minor issues.
• 5 — Excellent: High quality; clear impact and rigor.

Score calculation

• Raw per-dimension scores are stored in a JSON file and passed to:

  python scripts/calculate_scores.py --scores <path_to_scores_json>

• The script computes an aggregate score (and any configured weighting logic) based on the provided metrics.