Doc-to-Markdown Ingestion Pack for Make.com — Universal Converter, Knowledge Digest & Broken-File Guard Bots

Inspired by microsoft/markitdown, a document-conversion library trending on GitHub this week (184k+ stars) for turning PDFs, Office files, images, and audio into clean, LLM-ready Markdown: it preserves headings, tables, and links so the output is something an AI agent can actually read, instead of a wall of stripped text. Most teams don't want to run a Python library themselves — they just have a Drive folder full of PDFs and slide decks that no AI agent or knowledge base can actually use yet. This pack translates the same universal-conversion pattern into three scenarios you build in Make with tools you already have.

Included templates
- Universal Ingestion Router — watches a Drive/Dropbox/Gmail-attachment "Inbox" folder for new files, detects the file type (PDF, DOCX, PPTX, XLSX, image, HTML), routes each to the matching conversion path using Make's document modules plus an AI vision module for images and scans, normalizes the result into clean Markdown with headings, tables, and lists preserved, and writes the .md file into a "Knowledge Base" folder with source metadata — filename, type, page count, converted-at timestamp — logged to a Data Store
- Knowledge Digest Bot — triggers on each newly converted Markdown file, sends it to an AI module (OpenAI or Claude) for a structured summary: a three-sentence abstract, keyword tags, and a suggested category, writes the summary card into Notion or Airtable, and posts a one-line digest to Slack so the team knows what just became searchable
- Conversion QA & Broken-File Guard — checks every conversion output for red flags (a near-empty file, garbled OCR text, missing expected sections), and when a file fails, retries once through a fallback conversion path; if it still fails, routes the original file plus the specific error to a Slack or email review queue with an approve/retry/discard step — nothing gets silently dropped from the knowledge base

Every scenario follows the same pattern as its inspiration: convert conservatively, preserve structure, and never let a bad file pollute the knowledge base silently.

Common use cases
- Teams building an internal AI knowledge base or RAG pipeline who need documents in clean Markdown before they can chunk and embed them
- Ops teams drowning in PDFs and slide decks nobody has time to manually reformat
- Anyone piping files into an AI agent (Claude, GPT) who wants a reliable ingestion layer instead of pasting text by hand

Requirements
- A Make.com account (Core plan or higher, for scheduling and Data Store access)
- A Google Drive, Dropbox, or Gmail connection for the source inbox
- An OpenAI or Claude API connection for summarization and vision-based OCR fallback
- Notion, Airtable, or Google Sheets for the Knowledge Base and digest log
- Slack and/or Gmail for notifications and the review queue

What you receive
You get a detailed, module-by-module build guide plus a scenario map for each of the three bots — trigger, filter logic, conversion routing, AI prompt structure, Data Store/Sheet schema, router branches, and notification steps all specified. Make.com blueprint exports are account-specific (they embed your own connection IDs), so rather than shipping a blueprint file that would need reconnecting anyway, this pack gives you the exact scenario map to rebuild each bot in your own account in under 30 minutes per template — no guesswork required.