When Research Meets Reality: Building an AI Note-Taker (With Some Reluctance)

For my school thesis, I developed LectureSummarizer - an AI-powered application that records lectures, transcribes them using OpenAI Whisper, extracts keywords with TF-IDF, and generates study notes using local language models. While the final product works as intended, the journey taught me more about academic compromise than artificial intelligence.

The Academic Dilemma

The project emerged from a practical problem: some students struggle to take notes quickly enough during lectures. However, my literature review revealed something interesting - extensive research suggests that automated note-taking tools might actually be counterproductive for most students. The act of manual note-taking, despite its challenges, plays a crucial role in active learning and information retention.

Yet here's where academic realities intervened. Despite these research findings pointing toward the limited utility of such applications, building an AI solution had become somewhat of an institutional expectation. Sometimes you build what the curriculum demands, not necessarily what the research recommends.

Technical Implementation

The application combines several modern technologies:

• Audio Processing: Records live lectures using Python's sounddevice
• Speech Recognition: OpenAI Whisper for accurate transcription
• Keyword Extraction: TF-IDF for identifying key concepts
• Content Generation: Local GGUF language models for creating explanatory text
• Interface: Simple Tkinter GUI for ease of use

Research-Informed Design

Rather than ignoring my literature findings, I designed the tool specifically for "low efficiency listeners" - students who genuinely struggle with traditional note-taking methods. The application doesn't replace active engagement but provides structured support for those who need it most.

The software generates keyword-based summaries rather than comprehensive transcripts, encouraging users to still engage with the material while providing a safety net for important concepts they might miss.

Open Source Philosophy

I released LectureSummarizer under GPL v3, making it freely available and redistributable. If we're going to build AI tools for education, they should empower learners without creating new barriers or dependencies.

The complete source code, installation instructions, and documentation are available on GitHub.

Lessons Learned

This project taught me that sometimes the most valuable research outcome isn't the product you build, but understanding why you probably shouldn't build it - and then finding ways to build it responsibly anyway. The gap between academic requirements and research conclusions can be uncomfortable, but navigating that space thoughtfully can lead to more nuanced solutions.

For students considering similar projects: always let your research guide your implementation, even when external pressures suggest otherwise.