AI Tool

• Answers specific questions on scientific publications (~2500–3000 articles in total).
• Automates the process while keeping great results.
• Addresses optimization goals:
  • Reduce tokens used
  • Provide clear and better responses
  • Lower the hallucination rate
  • Get the most from AI
• Returns a pre-defined Excel document with Q/A columns.
• Used in a pharma research lab at CHU Sainte-Justine, Montreal to rework their website.

How to Use the Project

---

Step 1 – Download the Project

If you received this as a ZIP file, unzip it on your Desktop.

Right-click on the forlder, find 'Open in terminal'

Skip to next step.

If you’re using Git:

On your Desktop, right-click and find 'Open in terminal'

Once openned, run the commands :

git clone https://github.com/simy46/ImpactPharma.git
cd ImpactPharma

---

Step 2 – Add Your PDFs

Place all the PDF files you want to analyze inside the /pdfs folder of the project.

These are the pdfs that will be used : Chose the ones that YOU want.

Example: ImpactPharma/pdfs/*.pdf

---

Step 3 – Run on Windows

Make sure you have Python 3.10 or newer installed.

Then on the terminal

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
python main.py

---

Output

After the script finishes, results will be saved in:

• /outputs/resultats_YYYYMMDD_HHMMSS.xlsx : Excel with all extracted answers
• /logs/pipeline_YYYYMMDD_HHMMSS.log : detailed processing log for each PDF

---

Notes

• Don't edit the Excel file while the script is running.
• If you have questions or errors, share the log file with the developer.
• It will always write inside a new /outputs/resultats_YYYYMMDD_HHMMSS.xlsx with the date and time written on the name of the file.

Motivation

This is a work for a research lab at a Montreal hospital (the most popular if you ask me). I am happy to help them achieve their goal.

It is just a gpt-4 wrapper

I am no llm dev (yet : 10/07/2025), but I'm leaning towards that path more and more. I do love research and might get a paper out of this project.

https://impactpharmacie.org/index.php?p=greeter.php

Simple Approach (abstraction)

for pdf in pdf_files:
    text = PDFLoader.extract_text(pdf)
    responses = {}

    for category in categories:
        prompt = PromptManager.build_prompt(category, text)
        answer_raw = APIManager.ask(prompt)
        parsed = ResponseParser.parse(answer_raw)
        responses.update(parsed)

    ExcelWriter.insert_row(pdf_name, responses)