Finding anything is manual
One clause, one discharge note, one signed page. Somebody opens files one at a time until it turns up. Hours per question, and sometimes it never turns up at all.
Scan it once. Ask it anything. Always see the page the answer came from.
An entire paper archive, read inside your own building and answerable in plain language — without a single document being uploaded, shared or sent anywhere.
45 minutes, on your own documents. asor@asorahura.com
Courts, banks, hospitals and ministries hold decades of scanned and handwritten paper. The tools that could read all of it want the case files, customer records and patient notes sent somewhere else first — so the archives stay shut.
One clause, one discharge note, one signed page. Somebody opens files one at a time until it turns up. Hours per question, and sometimes it never turns up at all.
A system returns a table. Nobody can say which page it came from, so a regulator will not accept it, a court will not rely on it, and a clinician will not trust it.
Most approaches charge you to read the same pages afresh for every new question. The cost of being curious is what stops people asking.
A phrase you half remember is enough. No file reference, no knowing which cabinet. Results land in seconds, marked on the page.
Say what columns you want in a sentence. Rows come back across the whole archive, each field carrying a confidence score and editable in place.
The archive is already in. A question nobody thought of during scanning costs a fraction of the original pass, not a new project.
Every value carries a citation to the exact region of the original scan, and exports carry the citations with them.
Installed on hardware in your own building, or into your own cloud account if you prefer — the same system either way, and your choice to make.
Tap or hover any field to see the exact region it came from. This is what makes an answer usable as evidence instead of merely impressive.
Every field is checked back against the passage it was read from before it is shown to you, and scored. Anything that does not clear the bar is flagged for a person rather than quietly returned — which is why the order, the one field nobody should ever guess at, is the one being held back here.
Illustrative record. Fields and confidence scores shown are an example, not a real matter.
Discovery, precedent and case-file retrieval run on the same arithmetic: more paper than anyone can read, a deadline that does not move, and material so confidential it cannot be sent anywhere.
“anything that touches the easement over the access road” — every passage across the whole set, each opening on its own page, ordered by how squarely it answers. A narrowed review set on day one instead of week three.
“what did we argue last time a limitation point came up?” — the firm has argued it before; the difficulty is that it is in a matter file nobody remembers the name of.
“the 2016 haulage matter, claimant’s name started with O” — the file located from a partial recollection, with no suit number, and a record of who asked kept automatically.
“suit number, parties, date and orders for every appeal last year” — rows across the whole archive, each cell still linked to the page it was read from.
Privilege is preserved because nothing is ever transferred — to us or to anyone else.
The paper record holds the history that explains the patient in front of you — what was tried, what reacted, what was decided and why. It is in a box downstairs, so in practice it is not consulted.
“has this patient reacted badly to an anaesthetic before?” — the relevant admission surfaced out of three volumes and shown on its original page, handwriting and all, in the minutes actually available.
“every document relating to this patient between 2009 and 2014” — completeness is the legal test, so every page is accounted for and the release is logged immutably.
“discharge summaries mentioning infection after this procedure” — cases found by what the notes say rather than by whether anyone coded it, with the paper era included.
“what procedure is recorded in the operation note?” — the note itself on screen, with anything read from handwriting scored and the doubtful held back for a person.
No data-processing agreement to negotiate, because no patient data moves.
Examinations, remediation and disputes all end in the same place: legacy paper, a deadline set by someone else, and an obligation to produce the actual document rather than an assurance about it.
“every document supporting these facilities, including approvals” — the supporting file assembled across the archive, each item cited to its page, with the gathering itself logged.
“which of these accounts have no identification on file at all?” — what is genuinely held, read off the files rather than inferred from a system field, with gaps surfaced as gaps.
“the covenants and security position across this portfolio” — terms as executed, amendments and side letters caught, every value linked back to the signed page.
“what terms did this customer actually sign in 2013?” — the executed version distinguished from drafts, inside the complaint-handling window rather than after it.
Residency questions do not arise, because the material never moves.
Paper goes in at a resolution that holds up on small print, stamps, marginalia and handwriting — the parts of a record that matter and that cheap scanning destroys.
A single compact computer that runs from an ordinary wall socket and sits in the records room. No rack, no server room, no data-centre contract.
The whole platform on that one machine, operable by the person who already runs the records department rather than by a specialist team.
Because nothing goes out, the security review is short: a records department hosts its own document intelligence beside the filing cabinet it replaces.
And what it does with a pageFour decisions that were taken deliberately at the start, because none of them can be added afterwards.
Nothing is composed from general knowledge or filled in from what seems likely. If the archive does not contain it, what comes back is that it does not — which is the behaviour that makes the output safe to act on.
Each value is verified back against the passage it was read from and given a score. Below the threshold it is flagged for a human rather than quietly returned. A confident wrong answer is the only failure that really matters.
The link between a value and its region on the page survives every stage and comes out the other end in the export. It is not decoration added at the end — which is exactly why it cannot be bolted onto a system that was not built for it.
Accuracy is held to a fixed standard against a regression corpus, and a drop fails the build the way any other defect does. Over five hundred automated tests run against the platform.
These are internal targets, measured against our own corpus and held under test. They are not a published benchmark and not an independently audited result. Validation on your documents is part of a pilot — which is the honest place for a number like this to be settled, and the reason we would rather run the session on your paper than show you ours.
| Category | Runs unplugged | Citation to the page | Asking again | Burden on you |
|---|---|---|---|---|
| Cloud document AI | No | Partial | Billed again | Low |
| Enterprise AI search | Sometimes | Passage-level | Included | High — needs a specialist team |
| Legacy scanning & capture | Yes | Template-bound | Re-scan | Medium — rigid templates |
| Build it yourself | Yes | Build it | Depends | Very high |
| LocaLens | By default | Every field | A fraction of the first pass | One machine, one admin |
The honest competitor is a capable internal team with open tooling. The difference is the year they would spend making verification and citation actually hold under audit — and who gets called when it is wrong at the wrong moment.
The platform exists and is in production hardening. What follows is what is shipped, not a roadmap.
Bulk upload of whole archives, every page read including handwriting, with live progress you can watch rather than a job that disappears for a week.
Search across the entire archive that copes with a half-remembered phrase, scoped to one document, a folder, or everything at once.
Reusable templates for the tables you pull repeatedly, a citation on every field, re-asking without re-scanning, and inline correction where a human overrules it.
Organisation isolation, viewer/editor/admin roles, document-level permissions, and an immutable audit log that exports for an inspection.
Export to CSV, PDF, XLSX and PPTX, with the citations travelling alongside the values.
We turn up with the scanner and the machine, and leave with your archive answerable. There is no data-processing agreement to negotiate, because no data moves — which is what collapses the usual procurement cycle from quarters to weeks.
A single archive in a single building, chosen because you already know what is in it. That is the only honest way to find out whether it works on your paper rather than on a demo set.
The registrar, compliance officer or head of records who walks to the shelf today becomes the person who answers in seconds. They are the ones who will tell you whether it is real.
Immutable audit logs, role-based access, document-level permissions and nothing leaving the premises. It is the checklist your security reviewer already has, and it is what the system was built to.
Forty-five minutes, run on documents you recognise, with nothing handed over to make it work. Bring the file you would least like to have to find under a deadline.
The demo runs on a laptop with the wi-fi switched off. That is the pitch.