Skip to content

Blog

Documents, privacy and data for training models

What we learned building the PDF-to-dataset chain with no server, sources in plain sight. No fluff and no listicles.

datasetsfine-tuningprivacy
How to turn your PDFs into a training dataset without uploading them anywhere
From PDF to JSONL: extract the text, redact, drop duplicates, split by document and export in the format your trainer expects. All on your machine.
September 22, 20269 min read

datasetsevaluationmachine learning
Four mistakes that ruin a dataset built from your documents
Duplicates, contamination, leakage through chunking and personal data: four documented mistakes, how to fix each, and why the third one fools your metric.
September 22, 202611 min read
privacysecuritycomparisons
What happens to your PDF when you upload it to a free website?
What the public policies of the most used PDF sites say about how long they keep your file, and how to check yourself whether a tool uploads it.
September 22, 20268 min read
redactionChatGPTClaude
Redact a document before pasting it into ChatGPT or Claude
How to replace IDs, names and phone numbers with consistent placeholders, ask the AI with the anonymous text, and restore the real data in the answer.
September 22, 20267 min read
MCPagentsClaude Code
Give your AI agent access to your PDFs without uploading them: local MCP
What an MCP server is, why a local one lets Claude Code or Cursor work with confidential documents, and how to install it in one line.
September 22, 20269 min read
fine-tuningJSONLTRL
Dataset formats for fine-tuning: chat, Alpaca, ShareGPT and which one you need
The field names each trainer expects, verified against its official documentation, and the details that break a file without telling you why.
September 22, 202610 min read

Here for the datasets?

The niche page gathers the whole chain: from a folder of PDFs to an audited JSONL, without uploading anything.

See how it works

Every article has a Markdown version for agents and models: /llms.txt