NaDIA for engineering and geomatics firms
Tailored. Sovereign. Secure.
NaDIA brings your firm’s archive, ERP, drawings and standards together in a single assistant. Ask a question in Italian, by voice or in writing: it answers with your documents and cites the sources.
The firm’s knowledge lies in its archive, but it is hard to find.
Twenty years of projects sit side by side on the same network share, split into folders that each person has organised in their own way.
- Many PDFs are scans
- Digitised records, plans and correspondence contain no machine-readable text until they go through OCR.
- Emails matter
- Decisions, variations and agreements with clients and technical offices are often found only in archived email.
One archive, hundreds of projects
Number of files
A few million
about¼of the filesDocuments with text: about a quarter of the files.
Storage used
Tens of terabytes
under5%of the volumeDocuments with text: under 5% of the volume.
Windows search and file names are not enough.
They find the exact word, not the concept, and they do not read scans. They do not know that “assessment”, “inspection report” and “condition of the structure” are about the same thing.
So the answers stay in the memory of the firm’s most senior people.
One place for all your work.
Every firm has its own software and its own way of working. We do not ask you to change it: we adapt NaDIA to your workflows.
Today
separate tools
- Network archive
- ERP
- SQL databases
- CAD drawings
- GIS, INTERLIS
- Standards
- Site photos
With NaDIA
one place
- Network archive
- ERP
- SQL databases
- CAD drawings
- GIS, INTERLIS
- Standards
- Site photos
and others, made to measure
How the workflow changes.
| Topic | Before | With NaDIA |
|---|---|---|
| Search | Navigating endless folders and isolated databases. | Direct questions in Italian, spoken or written. |
| Formats | Opening a different program for each technical file. | Built-in viewers, right in the browser. |
| Answers | The firm’s institutional memory depends on individuals. | Every answer cites its sources, from your documents. |
Search
BeforeNavigating endless folders and isolated databases.
With NaDIADirect questions in Italian, spoken or written.
Formats
BeforeOpening a different program for each technical file.
With NaDIABuilt-in viewers, right in the browser.
Answers
BeforeThe firm’s institutional memory depends on individuals.
With NaDIAEvery answer cites its sources, from your documents.
Made to measure.
NaDIA makes the most of the tools you already have and adds new functions. It accesses every source read-only; other formats are added on request.
- ERP systemsBusiness Central
- SQL databasesRead-only
- Network archivesRead-only
- Email.msg and .eml, with attachments
- PDF and OfficeScans included
- DWG and DXFText, dimensions, layers
- Shapefiles and INTERLISOn the swisstopo map
- Photos and videosWith thumbnails
What NaDIA does in the firm.
The system is a website reserved for people in the firm, with the NaDIA assistant always available in a side panel. It is used in the browser, in the office, on site or on a phone, with nothing to install on computers.
Question, spoken or written
“What does the agreement with the municipality provide for the maintenance of the road?”
- Archive
- ERP
- Standards
NaDIA searches your connected archives, with your permissions.
Answer, with sources
It cites the exact source, from your documents.
Sources cited: PDF1MSG2DWG3
Search and the assistant.
You type as you would speak to a colleague. Typical time for a search across the whole archive: about 1.5 seconds2. The system combines three passes.
- 1
Exact words
Parcel numbers, codes and abbreviations stay whole and are found even when they are written only once.
- 2
Meaning
“Assessment” and “inspection report” are recognised as related even without a word in common.
- 3
Re-ranking
A second model rereads the top 50 candidates and puts the most relevant at the top.
- 01Results grouped by project, with project code, document type, highlighted excerpt and relevance on three levels.
- 02Filters by family (drawings, PDF, images, Office, email), by format and by year.
- 03The project code opens the project page; the chapter limits the search to one part of the archive.
- 04Each document opens in the viewer, can be downloaded or can be selected for NaDIA.
- 05An Excel file of the results with project, document, type, relevance and path.
- 06Recent searches stay below the field; the “History” page lets you run them again.
The text you search for never appears in the page address: it is not left in logs or in the browser history.
NaDIA works on the firm’s archive, not on a model’s memory: it searches, counts, reads the relevant documents and only then writes. The model never queries the databases directly.
Content
“What does the agreement with the municipality provide for the maintenance of the road?”
Reads the 5 most relevant documents and cites them.
Summaries
“Summarise project 2024-117 for me.”
One document or an entire project.
Comparisons
“Compare the deadlines in the two specifications.”
Up to 8 documents chosen from the results.
Counts
“How many quotes from 2025 are there?”
The number comes from the catalogue, not from an estimate by the model.
Drafts
“Write me a note for the client with the documents found.”
A Word file marked as a draft to be verified, with the sources.
Navigation
“Take me to the 2023 DWG drawings in chapter 05.”
Sets the filters and opens the results.
Project lists and figures read from the ERP come as tables, with “Download as Excel”.
- Art. 1
The system, not the model, cites the sources.
Beneath each answer, the documents and queries actually used are listed.
- Art. 2
No invented figures.
An answer with figures that are not in the sources is discarded. The same applies to Word drafts.
- Art. 3
“The sources do not cover this” is an accepted answer.
Below a relevance threshold NaDIA says so, instead of improvising.
- Art. 4
NaDIA changes nothing.
Its tools are read-only, and the content of documents is treated as data, never as instructions.
From question to answer, on the same scale in seconds. These are typical timings2.
Typical timings
A search across the whole archive: about 1.5 s
Transcribing a dictated sentence: about 2 s
A navigation or a count: from a few tenths to 8 s
A Word draft: about 12 s
Reading and comparing two or three documents: between 25 and 30 s
Voice
A microphone in the NaDIA panel records up to 60 seconds, on a computer or a phone, even on site. The transcription is sent straight away as a question. Typical time for one sentence: about 2 seconds.2
History
Conversations reopen where they left off. NaDIA also finds its own conclusions: “what did we conclude about the bridge assessment?” Everyone sees only their own.
Attachments
Up to 5 files per message, 20 MB each. A PDF of up to 300 pages is read as text; scans and photos as images. For example, a contractor’s bid to compare with the project specification.
Every evening
A document saved today can be found tomorrow morning, a scanned PDF the day after. If an abnormal share of files changes or disappears in one evening, the system stops and leaves the index untouched.
What it reads.
In a technical archive, scans can account for up to a third of the PDFs. OCR, run locally, makes almost all of them readable: in the order of 1% of PDFs remain unreadable, and these can still be found by project, folder and name. Other formats are added on request.
The PDFs of a technical archive
- PDFs with text: at least two thirds
- Scans, local OCR: up to a third
- Unreadable: about 1%
The text becomes searchable, including the text of scans.
PDFs with text
PDF
Searchable: All the text
In the browser: Yes
Scanned PDFs
PDF
Searchable: The text recognised by OCR, locally
In the browser: Yes
Word, Excel, PowerPoint
DOCX, XLSX, PPTX, DOC, XLS, PPT, RTF, ODT
Searchable: Text, tables, slides and notes
In the browser: Yes
Email
MSG, EML
Searchable: Headers, body and attachments with text
In the browser: Yes, with attachments that open
Text, dimensions, objects and attributes, with the drawing in the browser.
CAD drawings
DWG, DXF
Searchable: Text, dimensions, layers and blocks
In the browser: Yes, zoomable drawing
Shapefiles and KML
SHP, KML, KMZ
Searchable: Attributes and names
In the browser: Yes, on the swisstopo map
INTERLIS
ITF, ILI
Searchable: Objects and attributes; models as a tree
In the browser: Yes, clickable objects
Catalogued even when they have no text.
Photos
JPG, PNG
Searchable: Name, folder, date taken
In the browser: Yes, with thumbnails
Point clouds, rasters, videos, archive files
Searchable: Name, folder, date, size
In the browser: Network path; videos in the viewer
Folders, viewers and permissions.
The system knows every file on the share: point clouds, photos, videos and archive files too, which take up most of the space in a geomatics firm.
Project page
For each project: the folder tree, totals by file type, photo thumbnails. Files below a certain size open or download; for larger ones, such as a 5 GB point cloud, the network path appears with “Copy path”.The “Folders” page
Two panes, as in File Explorer: the tree on the left, the contents on the right, sortable by name, date and size. The firm’s archive can be navigated in the browser, even away from the office.
Viewers in the browser
Files open over the page, without being downloaded and with no software to install: PDF and Office with sheets, slides and notes; emails with attachments that open; DWG and DXF with zoom; shapefiles and KML on the swisstopo map; INTERLIS with the codes translated into words; photos and videos.Photo survey
“The photos of project 2024-117 taken in May”, then “make a photo survey of them”: out comes an A4 Word document with two photos per page, grouped by folder, with number, date and a “Description” field to fill in.
Administrator
In addition: user management, usage and price list, permissions per person
Manager
In addition: archive status, activity log, NaDIA quotas per person
User
Search, Folders, Standards, NaDIA, History
Restricted folders
The administrator can restrict top-level folders, for example the staff folder, and grant access person by person. For anyone without access, the folder does not exist anywhere in the system: search, Folders, NaDIA, counts, exports.
Monthly spending cap per person
Every use of AI is recorded. At 80% of the cap a warning is issued; at 100% NaDIA stops for that person, while search continues.Archive status
How many files are already searchable, how many are being processed or waiting for OCR, the pace, the estimated finish and the next run.Activity log
Logins, files opened and downloaded, changes to users and permissions, reads of the ERP. Never the content.
Standards and ERP, in the same place.
NaDIA never says from memory whether a law or standard is in force: the status comes from the official source, with the date of the data. Every night the system updates the list of federal laws from Fedlex, about 13’000 acts, and the public list of SIA standards, about 2003.
- Finds a law or standard by its abbreviation, SR number, “SIA 405” or words in its title. It shows whether it is in force or since when it has ceased to be, what replaced it, and links to the Fedlex page.
- Finds the right law even without its abbreviation, with a search by meaning on the titles.
- Checks a pasted report or specification: the table of the laws and standards cited shows the status of each, superseded or repealed ones first.
- In the chat you ask “is the SIA 118 cited here the current edition?”, and NaDIA checks the laws and standards in a document from the archive, an attached file or the selected documents.
The limits of the current version
- Cantonal and municipal laws are not covered.
- It recognises the law, not the individual article.
- Of SIA standards it knows the number and edition, not the text, which is protected by copyright.
The ERP, read-only.
For firms that use Microsoft Dynamics 365 Business Central, NaDIA also answers questions about the figures in the ERP.
- Free-form questions to NaDIA: “overdue unpaid customer invoices”, “supplier invoices due in the next 30 days”, “offers still open”.
- The figures are read live and read-only: no copy of the ERP data remains in the system.
- Totals are calculated by the system, not the model; the query that was run appears in plain text under the table.
- The administrator decides who can query the ERP; employees’ personal data is excluded.
For those who prefer filters, the “ERP” page: accounting, quotes, orders, projects and hours.
- tables, read-only
- 51
- filters at most
- 5
- rows exportable to Excel
- 5’000
In the current version the ERP does not provide costs, revenue and margins per project: it depends on how they are recorded in Business Central, and needs to be checked against the firm’s data.
Your data stays yours. Your tools work together.
Protection. Integration. Intelligence.
The archive stays in the firm. The artificial intelligence, in Switzerland.
The system accesses the archive read-only and does not copy it elsewhere. Extracted texts, indexes and conversations stay on a server inside the firm.
Protected and connected.
Protected
Designed for the firm’s most sensitive data.
- 01
Security by design
No open inbound port, an encrypted connection, the archive accessed read-only.
- 02
Sovereign artificial intelligence
Model and AI services on servers in Switzerland. Your documents train no model.
- 03
FADP and GDPR by design
Only the extracts needed leave the firm; retention decided by you; a log of every action.
Connected
One place for what is scattered today.
- 01
Tailored profiles and permissions
Roles and permissions for users and groups, down to the individual folder: everyone sees what they are entitled to.
- 02
Your ERP systems and databases
ERP systems, SQL databases, network archives: NaDIA reads them live and read-only.
- 03
Viewers built into the browser
PDF, Office, email, DWG, shapefiles, INTERLIS, photos and videos, with no software to install. Other formats on request.
The website only passes data through.
The system has two parts: a server in the firm, which reads and indexes the archive, and a website on servers in Switzerland, which people access from the browser. The two parts communicate through an encrypted tunnel opened from the firm, so the firm’s firewall opens no inbound port.
In the firm
- Project shareRead-only
- Extraction and OCRAt night
- Texts, indexes and conversationsOn the server’s disk
- Search serviceOnly in the tunnel
- Text fragments, questions, voice
- Encrypted tunnel, opened from the firm
In Switzerland
NaDIA’s AI servicesSemantic index, re-ranking, voiceNaDIA website, run by Native DigitalLogins, permissions, historyPeople in the firm: from a browser or a phone, over HTTPS.
- Outgoing calls, only with the extracts needed, neither stored nor used for training
Services called by the website
- NaDIA’s model, in Switzerland
- Fedlex and SIA list: status of laws and standards
- Business Central: read-only
In short: where the data is.
- Where the documents are
On the firm’s network share, accessed read-only.
Extracted texts, indexes, conversations and attachments on a server inside the firm.
- Where processing takes place
The heavy work (OCR, extraction, indexing) on the firm’s server.
The website (accounts, permissions, history) on a server in Switzerland run by Native Digital. NaDIA’s model and AI services in Switzerland.
- What goes out
Only the text fragments and extracts needed to search or to answer,
together with questions and voice audio, sent to NaDIA’s services in Switzerland.
- What is neither stored nor used for training
Those fragments and extracts are neither stored nor used for training.
The website writes none of the firm’s files or text to its own disk.
Item by item: where it is and who sees it.
In the firm
Original archive files
Where it isThe firm’s network share, accessed read-only
Who sees itAs today
Extracted texts, indexes, catalogue, previews
Where it isThe system’s server, in the firm
Who sees itOnly through the website, according to permissions
Conversations with NaDIA and attached files
Where it isThe system’s server, in the firm
Who sees itThe owner, through the website; the server administrator
In Switzerland
Accounts, search history, usage, log
Where it isThe website, on servers in Switzerland
Who sees itThe owner; usage and log to managers and the administrator
Text fragments to be indexed, questions, voice audio
Where it isProcessed by NaDIA’s services, in Switzerland
Who sees itNo one: they are neither stored nor used for training
Extracts, attachments and ERP rows needed to answer
Where it isProcessed by NaDIA’s model, in Switzerland
Who sees itNo one: they are neither stored nor used for training
For your IT manager.
No open port into the firm
The connection to the website is an encrypted tunnel opened from inside the firm, limited to the search service alone.The website only passes data through
None of the firm’s files or text is written to the website’s disk; audio and attachments stay in memory for the time needed.Access
Accounts created only by the administrator; passwords of at least 12 characters, hashed with Argon2id; progressive lockout after 5 attempts; session closed after 30 minutes of inactivity.Connection
Mandatory HTTPS, protected cookies, defences against forged requests and external scripts.Read-only ERP
Read access only, to a closed list of tables. We recommend granting access through a dedicated application, with read-only permissions on the Microsoft side as well.Secrets and backups
Keys and passwords encrypted and bound to the machine; an encrypted nightly copy of the system’s database, to sit alongside the firm’s backup plan.
- Server in the firm
- Windows Server, physical or virtual; 8 virtual processors; no graphics card
- Memory
- 64 GB minimum; 96-128 GB for an archive of a few million files
- Disk
- SSD of at least 500 GB, 1 TB recommended
- Network
- Outbound TCP to the website and HTTPS to the services; no inbound port
- Data access
- A read-only account on the project share
- Website
- Provided and run by Native Digital, on servers in Switzerland
- Workstations
- Nothing to install: a recent browser, including on a phone
- Backups
- Backup space off the machine, within the firm’s plan
The memory is used by the semantic index and grows with the archive: about 7 GB for every million text fragments. The initial inventory makes it possible to size it to your figures.
How to get started.
Installation does not stop work: the archive is accessed read-only and the heavy processing runs at night.
1
We listen
We study how the firm works, to understand your needs.
2
We connect
We bring together your archive, ERP and existing programs, without interrupting your work.
3
We refine
We adapt everything until everyday use comes naturally to the people in the firm.
Six weeks, from inventory to service.
Indicative timings: two weeks of configuration, then a month of trial use by the firm4.
The timings depend above all on the number of scanned PDFs. Then, every day: an automatic evening update, monitoring of the processes, updates and user support.
Week
Week 1
Inventory and start-up
Composition and volumes, OCR tests on a sample, server installed.
Week 2
Indexes and website
Recent years first, then the historical archive and OCR; website, roles and folders.
Weeks 3-6
Trial and fine-tuning
The firm uses the system in its everyday work; we fine-tune it to the firm’s requests and train the staff.
- Trial starts in week 3, in service after week 6.
What is needed from the firm.
Access, a machine and a few decisions. We do the rest.
- 01A read-only account on the project share.
- 02A dedicated Windows Server machine, physical or virtual, with the resources stated in the requirements and administrative access for the installation.
- 03Outbound internet access over TCP and HTTPS; no inbound ports opened.
- 04The list of people and roles, and any folders to be restricted.
- 0530-40 real questions that colleagues ask the archive today, to measure and calibrate the search on your documents.
- 06Your choices on data: how long to keep the history.
- 07For the ERP, if used: a dedicated read-only Microsoft application and confirmation of the Business Central licences for the people who will see the data.
For our part, we adapt the system to the structure of your folders (how project, code and chapter are recognised), to the firm’s name and image, and we prepare the user guide.
What does the current version not cover?
- Entire mailboxes (.pst) and the contents of .zip archives.
- Searching photos and drawings by their content.
- Cantonal and municipal laws.
- Costs and margins per project from the ERP.
NaDIA in the other sectors
All sectors
Next step: a read-only inventory of your archive, in a single day.
It gives you the composition, the volumes, the share of scans and an estimate of the time to go live.