Land certificates filed in 2078, including the ones held only as photographs of the original.
Why it matches
Scanned Devanagari pages are read on arrival, so text locked inside an image is indexed like any other document.
Find a record from a half-remembered phrase, a reference number, a date or a name, in Nepali or in English.
Almost nobody remembers the filename.
They remember a phrase from inside the document, roughly when it arrived, who sent it, or what it was about. A folder tree answers none of those. Each document is prepared on arrival so that all of them work as ways in.
Text from digital files and from scanned or photographed pages is indexed, so a phrase buried on page four is as findable as the title on page one.
Category, document type, owner, branch and status, plus the dates, reference numbers and parties extracted on arrival, all become things a reader can search on.
Exact phrases, misspellings and near-misses, and plain-language descriptions of what the reader is after are all accepted, in Nepali and in English.
The result set is cut to what this particular reader is allowed to see, so a restricted record never appears to anyone else, not even as a title.
Every result carries the passage that matched alongside its category, status and owner, so the right one can be picked without opening four of them first.
Real queries, what each one returns, and the reason it works. None of these depend on the reader knowing where the document was filed.
Land certificates filed in 2078, including the ones held only as photographs of the original.
Why it matches
Scanned Devanagari pages are read on arrival, so text locked inside an image is indexed like any other document.
The agreements with that counterparty and their end dates, despite the misspelling.
Why it matches
Near-misses and typos are tolerated, so a mistyped query narrows the result set instead of emptying it.
The submission itself and the email thread it went out on, with the answering passage shown.
Why it matches
Email is indexed alongside documents, so a decision and the correspondence that carried it are retrievable together.
Current minutes from the period, with superseded and draft versions kept out of the way.
Why it matches
Status and date are fields on every record, so a filter does the narrowing rather than the reader scanning a long list.
In each of these the document already exists and is authoritative. What is missing is a path to it that does not depend on remembering how it was filed.
A file saved as scan0032 or Final_v3 is invisible to anyone who did not save it. Indexing the contents rather than the label means the document answers to what it says.
Operations files it as a vendor agreement and finance as a supplier contract, so neither finds the other’s copy. Searching the contents and the extracted parties gets to the same record from either vocabulary.
The request is for whatever supports a figure or an approval, wherever it happens to sit. Searching text, extracted fields and indexed email together produces the supporting documents as a set.
A customer is waiting while someone calls head office to ask which rule applies. Permission-scoped search over the shared record set lets the branch find the current circular themselves.
Yes. Scanned and photographed pages are read automatically as they arrive, in Nepali and English, and the resulting text is indexed alongside everything else. A phrase inside a photographed register is findable the same way a phrase in a Word document is. Handwriting is the exception and is not yet supported.
Near-misses and typos are handled rather than returning an empty result set, which is the failure mode that teaches people to stop using search. Names transliterated between Nepali and English are the common case, and a query that is close enough still reaches the record.
No. Permissions are applied before results are ranked, not after, so a restricted record is absent from the result list entirely rather than shown and then blocked. It does not appear as a title, a snippet or a count, which means search cannot be used to work out what exists.
No. Connected Google Drive, Dropbox and OneDrive folders and indexed email are searchable where they already live, so the first useful result usually predates any migration. Bulk import driven from a spreadsheet is there for when you do want to bring an archive in properly.
A shared drive matches filenames and, at best, the text of files it can already read. It cannot read a scan, does not know a document is superseded, has no idea who the parties are, and shows a hit without telling you why it matched. The difference is less about the matching and more about what has been indexed and what each result tells you.
Yes. File type, category, document type, owner, branch, status, date and source can all be applied on top of a query. Filtering to active documents only, or to a single branch, is usually the fastest way from a large result set to the specific record in hand.