OCR + AI title-block extraction
Stop hand-naming scanned drawings.
Nectar reads the title block on every sheet of a scanned set: sheet number, title, project, date. Then it renames and files the whole batch to your standard. A day of data entry becomes minutes of review.
Riverside Medical Center · Addendum 02
Credits 1,412
6 sheets · renamed and filed
⌘SRename
File
Sheet
Title
Status
A-101 - First Floor Plan.pdf
A-101
auto
FIRST FLOOR PLAN
auto
Done
A-201 - Exterior Elevations.pdf
A-201
auto
EXTERIOR ELEVATIONS
auto
Done
S-301 - Foundation Details.pdf
S-301
auto
FOUNDATION DETAILS
auto
Done
E-101 - Lighting Plan, Level 1.pdf
E-101
auto
LIGHTING PLAN, LEVEL 1
auto
Done
M-401 - HVAC Plan, Level 1.pdf
M-401
auto
HVAC PLAN, LEVEL 1
auto
Done
P-102 - Plumbing Risers.pdf
P-102
auto
PLUMBING RISERS
auto
Done
- Hundreds of sheets per drop
- Add a whole set at once. Uploads run in parallel, survive an app restart, and retry per file.
- Only uncertain sheets wait for you
- Confident reads go straight through. Anything below the confidence bar waits for a quick human check.
- Gone from our cloud in 72 hours
- Your PDFs are processed, then purged. We keep the extracted fields, not your drawings.
The pipeline
How Nectar reads a sheet.
Five stages between a scanner dump and a filed set. Most sheets never need you at all.
01
Drop the scans
Drag in the whole set, hundreds of files at once. A half-finished batch picks up where it left off after a restart, and any failed upload retries straight from the original on disk.
Ingest
01 / 05
Drop 214 files to upload
scan_0041.pdf
scan_0042.pdf
scan_0043.pdf
scan_0044.pdf
+ 210 more
parallel uploads · resumes after restart · per-file retry
02
High-resolution OCR
Nectar runs every page through high-resolution OCR tuned for the small text crammed into title blocks. You get a searchable PDF copy of every sheet as a free byproduct.
OCR
02 / 05
high-res mode for small title-block text · + searchable PDF copy
03
AI reads the title block
A fast text model reads every sheet first. Only the hard ones escalate to a vision model that looks at the title block itself. All of it runs inside our own cloud, and none of it touches a third-party AI service.
Extraction
03 / 05
SHEET NO.
E-101
SHEET TITLE
LIGHTING PLAN, LEVEL 1
text pass · every sheet
→
vision pass · only when unsure
04
Deterministic guards
Geometry beats guesswork. Nectar snaps every read to what's actually printed on the sheet. Discipline prefixes, rotated stamps, and the set's own drawing index keep the model honest.
Guards
04 / 05
-
snap: sheet number matches the corner-most label
-
guard: E- prefix agrees with the electrical discipline
-
rotation: 90° stamp detected, labels read in place
-
cross-check: title agrees with the set's drawing index
05
You review only the uncertain sheets
Sheets that clear the confidence bar are done. The rest queue up for a side-by-side check that takes seconds. And if Nectar got one wrong, reporting it refunds the credit automatically.
Review queue
05 / 05
E-101
LIGHTING PLAN, LEVEL 1
Review
wrong read? report it and the credit comes back
The review loop
Built for the person doing 400 sheets before lunch.
The viewer snaps straight to the title block, the extracted fields sit beside it, and the whole confirm loop runs from the keyboard. Bluebeam and Acrobat muscle memory carries over.
- ⌘Enter
- Accept & next. The whole review loop in one chord, without reaching for the mouse.
- Enter
- Confirm a field and move to the same field on the next sheet.
- ⌘↑↓
- Jump between sheets without leaving the field you're in.
- ?
- Every shortcut, on screen.
scan_0044.pdf
click any OCR line to fill the field
LIGHTING PLAN, LEVEL 1
RIVERSIDE MEDICAL CENTER
E-101
Sheet number
E-101
Sheet title
LIGHTING PLAN, LEVEL 1
⌘EnterAccept & next
The output
Your files, named your way.
Build the naming template once by dragging the pieces into order. Every set after that comes out to the same standard.
Filename template
{prefix}
_
{sheet_number}
-
{sheet_title}
007_E-101 - Lighting Plan, Level 1.pdf
008_E-201 - Power Plan, Level 2.pdf
collisions de-duplicated · names sanitized for Windows and macOS
- Renames in place
- Nectar renames your originals right on disk. It never copies or re-saves them, and it never touches their contents.
- Exports searchable copies
- A managed folder of OCR'd, text-searchable PDFs named by the same template, with viewer rotation baked in.
- Goes hands-off when you're ready
- Trust a set? Turn on auto-accept for high-confidence reads, auto-rename, and auto-export.
Security & data residency
Built to pass your client's security review.
You're handling other people's drawings. Here's exactly what happens to them, written so you can paste it straight into the IT-approval email.
- In-network AI
- Extraction runs on AI models hosted inside our own cloud account. We never send your documents to a third-party AI vendor, and we never use them to train models.
- Purged within 72 hours
- Uploaded PDFs are deleted from our storage within 72 hours of processing, with a hard 30-day backstop. We keep the extracted data, never your drawings.
- US and Canadian data planes
- Nectar runs dedicated environments in the United States and Canada, and your account is pinned to one. Documents, OCR, and the document database never leave the country. Need Europe? Reach out and let's talk.
- Non-destructive by design
- Nectar renames in place and exports copies. It never deletes files from your disk and never rewrites a drawing's contents. Installers are signed and notarized, with cryptographically verified auto-updates.
FAQ
Common questions.
Something else on your mind? Ask it in the beta request and we'll answer directly.
How does pricing work?
Entirely on usage. Credits are the only thing you buy, and one credit covers one file end to end. You pay for what you process, with volume discounts as your sets get bigger. Tell us about your volume in the beta request and we'll suggest the right starting point.
What happens when it reads a sheet wrong?
You fix it in seconds. The sheet and its fields sit side by side, and clicking any OCR line drops that text straight into the field. Report a bad extraction and Nectar refunds the credit automatically.
What kinds of sets does it handle?
The messy real-world ones: large-format scans, as-builts, sheets that are rotated, skewed, or hand-stamped, and non-standard title blocks. That's the corpus it's built and tested against.
What happens to my files?
On your machine, Nectar renames originals in place, writes exports as separate copies, and never deletes anything. In our cloud, processing copies are purged within 72 hours.
Which platforms does it run on?
Native desktop apps for Windows (x64) and macOS (universal). Installers are signed and notarized, and updates install themselves silently after cryptographic verification.
How does licensing work?
Per-seat license keys, one machine per seat. You manage seats yourself in the customer console: add one, release one, or move a seat to a new computer without emailing anyone.
Can my data stay in-country?
Yes. Nectar runs dedicated environments in the United States and Canada, and your account is pinned to one of them. Documents never leave the country they're processed in. Mention residency in your beta request and we'll set your account up in the right one. And if you need a region we don't run yet, ask. We're open to adding more.
Beta access
Put your next set through it.
Tell us about your shop and the sets you process, and we'll send a license key. We usually reply within one business day.