Media Summarizer

The problem

Processing hours of video, audio, and long-form articles for research or meeting prep is time-consuming. Manual note-taking from transcripts often results in missing key figures or including unsupported claims.

What it does

• Probes URLs to classify media types and estimate word counts before processing.
• Extracts text from web pages, PDFs (with OCR support), and transcribes audio or video files.
• Chunks large datasets on paragraph boundaries to maintain context during LLM analysis.
• Generates structured summaries with bullet highlights, key statistics, and a final verdict.
• Verifies summaries against source text to ensure a high claim-coverage ratio.

Frameworks & tools

Python-based CLI tool medsum, supporting LLM backends like GPT-4o-mini.

Why this beats prompting it yourself

Standard prompts often truncate long inputs or hallucinate details from large transcripts. This skill handles the heavy lifting of media extraction, context-aware chunking, and automated verification to ensure every bullet point is grounded in the source text.

Use cases

• Convert technical YouTube presentations into structured research briefs.
• Extract key decisions and action items from long podcast episodes or meeting recordings.
• Summarize batches of PDF whitepapers into a standardized internal knowledge base.
• Audit external summaries for accuracy by running verification against raw transcripts.

Known limitations

Does not support sources behind login walls. Recordings over three hours should be processed in segments rather than a single pass. Requires the medsum CLI installed locally.