ClearPathAI

A study tool that generates two summaries from the same passage and checks the ADHD-friendly version against ten readability and structure rules.

  • Python
  • Streamlit
  • Generative AI

Project overview

ClearPathAI turns dense study material into notes organized for learners with ADHD. The Streamlit app produces a baseline summary and a structured version from the same passage. The structured one organizes the material into learning objectives, key concepts, and recall questions.

I led evaluation and documentation for the team. I recorded how the two versions scored on readability and structure, reran ten sample passages three times each, and documented file-upload issues I found while testing.

How the summary generator works

  1. Provide study materialPaste a passage into the app.
  2. Generate two summariesThe model writes a baseline and a structured version from the same passage.
  3. Check and compareReview both summaries, their readability, and the results of each rule check.
The generator can make up to three attempts to produce a structured summary that reaches the 80% threshold.

What the rules measure

The Python checker scores ten rules equally. A summary passes the project’s threshold when it meets at least eight of them.

  • Length and readability

    The checker looks for sentences of up to 20 words, paragraphs of up to 80 words, and a Flesch–Kincaid grade level of 8 or below.

  • Structure and recall

    It looks for three named sections, blank lines between them, three to six key-concept bullets, and at least two recall questions. Formulas have their own lines.

What the evaluation showed

Across ten passages and three runs per passage, the baseline summaries scored 20–40% on the project’s rules. The structured summaries scored 90–100%. In the first run of the Python textbook passage, the Flesch–Kincaid grade level fell from 10.9 to 4.5. These measures describe the text’s structure and readability.

I pasted the passages for every scored run to keep the input method consistent. In a separate upload check, TXT generated summaries successfully, while PDF and DOCX hit errors. I documented those failures for follow-up.

Sources