Common foundation
Finding anything
A single index for files, conversations, artifacts and projects, a search that says where it looked, and archives you read without ever moving them.
The problem
The document exists somewhere: on the disk, in the shared drive, attached to an email, in an old conversation, in an artifact or in a presentation from three years ago. But you no longer know where, and searching takes longer than starting again. So you start again, and a fourth version of the same slide joins the other three. When you ask the AI, it answers “not found” without saying where it looked.
Step-by-step method
- A unified index covers files (local disk and shared drive), the conversation export, artifacts and projects. It has two parts: a public part that can be published safely, and a private part loaded locally on your computer.
- The conversation export stays in the browser: it is loaded into the page’s local storage and never leaves it. Sensitive topics appear only by title and date, never by content.
- Editable synonym groups: an acronym also finds its expanded form, its English or French equivalent and its spelling variants. Aliases cover the words that voice dictation tends to mangle.
- A search bar in the dashboard: it searches its own data first, then hands over to the full engine. The answer comes in three blocks: discussions, documents, related items. Each discussion is linked to its documents through the names of its attachments.
- A search with four statuses for each source: queried with results, queried without results, inaccessible, not queried. “Not found” is only said once the whole scope has been covered, and the answer states the age of the index and what it cannot see (folders not indexed, date of the last conversation export).
- Reindex only when something has changed: a fingerprint of the folders is enough to tell. The page is republished only above a threshold of changes.
- For presentations, a version inventory:
- versions are grouped by file name (V, VF, dates, copies), then by the title of the first slide;
- the same content less than 60 days apart counts as a single edition;
- the final and the longest versions are shown; the others are hidden but accessible, and each year of the same course stays visible;
- the search also covers the text of the slides;
- nothing is moved.
- Archives stay read-only:
- a map and an index first, before anything is opened;
- some folders are never opened without a request for a specific file (health, imaging, named reports, finances);
- work is always done on a copy suffixed with the date;
- duplicates are listed as a proposal, without moving anything.
Deliverable
A single search page, and answers that always say what was searched, where, and with what result for each source.
Safeguards
- No conversation export in a public online space: it stays in the browser on your computer.
- Real file names may contain patient names: never take a screenshot of the raw index or share it.
- Search reads, it does not file: filing goes through Organizing your files for AI.
Starter prompt
I am looking for [what I am looking for, with the words I remember]. Search the file index, the conversation export, the artifacts and the projects, widening to synonyms. Answer in three blocks (discussions, documents, related items), then give the status of each source: results, no results, inaccessible or not queried. State the age of the index and what it cannot see. Do not open any sensitive archive folder and do not move anything.