Browse all technical articles on the MDisBetter blog.
Signal-to-noise ratio, microphone choice, room treatment and pre-processing: what actually moves transcription accuracy from 60% to 99% on real recordings.
TechnicalArchitecture for migrating thousands of Word documents to Markdown: audit, categorise, prioritise, convert with Pandoc, quality-check, organise and publish.
TechnicalEngineering lessons for building a PDF to Markdown converter: pipeline design, heuristics versus models, OCR preprocessing, tables, speed, output format.
TechnicalVoicemails, meetings and interviews sit unindexed. Convert them to Markdown, organize by date and speaker, then search with ripgrep or Obsidian in seconds.
TechnicalTurn videos into a searchable text archive: transcribe them, organize the Markdown with frontmatter metadata, then search it all with ripgrep or Obsidian.
TechnicalEnd-to-end architecture for turning web sources into a queryable AI knowledge base: conversion, chunking, embedding, vector storage and update strategy.
TechnicalStatic fetch vs headless browser, Playwright and Puppeteer mechanics, wait conditions, and cost tradeoffs behind URL-to-Markdown tools on JS-rendered SPAs.
TechnicalFrom HMM-era speech recognition to encoder-decoder transformers and Whisper's 680k-hour training set, plus why structured Markdown output matters afterward.
TechnicalA .docx file is a ZIP archive of XML. Walk through document.xml, styles.xml, and OOXML to see why naive text extraction loses heading semantics.
TechnicalA technical deep-dive into the PDF file format: content streams, glyph positioning, why extraction is lossy, and what this means for AI workflows.
TechnicalHow YouTube captions work: auto-generated ASR versus creator-uploaded tracks, why auto-captions are unreliable, and how fresh AI re-transcription compares.
TechnicalTechnical deep dive: DOM parsing, tree-walking, element-by-element conversion rules, and why naive html2text falls short on modern web pages.
TechnicalTechnical comparison of three Word-to-Markdown approaches: Mammoth.js, Pandoc, and AI-powered conversion. How each works, its tradeoffs, and when to use it.
TechnicalWhy the same web page costs far fewer tokens as Markdown than as raw HTML: compact syntax, stripped chrome, and how to check the gap on your own pages.
TechnicalWhat plain-text transcripts lose and what Markdown preserves: punctuation, sections, timestamps, emphasis, and the LLM extraction quality difference.
TechnicalSRT and VTT are subtitle formats for video players, plain text is unstructured, and Markdown adds structure plus AI-readiness. When to use each format.
TechnicalThree chunking strategies for RAG pipelines: header-based, token-based, paragraph-based. When each wins, with code examples and evaluation metrics.
TechnicalA step-by-step method to count tokens on your own PDFs, their plain-text extract and their Markdown version, then turn the gap into a cost estimate.
TechnicalTechnical guide to integrating video into RAG: transcript to Markdown, chunk by H2/H3, embed, retrieve. Python with sentence-transformers and ChromaDB.
TechnicalA technical look at the main-content extraction problem. Mozilla Readability, Trafilatura, and LLM-based extraction compared: strengths and weaknesses.
TechnicalWord tables allow nested tables, merged cells, and multi-row headers. Markdown tables are flat rows and columns. What survives, what breaks, how to bridge it.