Extract data from vehicle titles
Unstract reads certificates of title, title pledge forms and lien releases from dealer scans and phone photos, confirms the VIN, owner and your lien, and sends each vehicle's title record to your loan origination system.
lien noted, ok to fund — MP
Specimen title · fictional state and owner
{ "title_number": "NR2508814736", "vin": "4N7BX3KD8PM216047", "year": "2023", "make": "Vanterra", "model": "Ridge LX", "odometer": "18,402", "owner": "Daniel K Okafor", "first_lienholder": "Cedarline Federal Credit Union", "lien_date": "03/14/2025", "issue_date": "03/28/2025"}
Trusted by lending and onboarding teams around the globe
Titles come from fifty DMVs, dealers and borrowers' glove boxes
Every state prints its own title, and the paper reaches the lender through a dealer's funding package, a DMV mailing or a borrower standing at the counter with a phone.
Titles photographed on a car seat, folded through the VIN line
One lender, three labels across three states' titles
Title pledge forms filled in by hand at the store counter
A dealer deal jacket scanned as one long PDF
Title front and the reassignment on the back, sent as two files
Title clerks key each VIN and lienholder name from the paper and chase dealers for titles that never show the lender, so loans fund with no lien perfected on the car that secures them.
Unstract reads the title from any state, extracts the VIN, owner, odometer and lienholder, checks them against the loan, and records the result in your loan origination system.
From a dealer's title scan to a perfected lien
Read the title, extract the vehicle and lien data, then let the run confirm your lien before the loan file closes. Each product works by itself, and the three together keep a car off the books until the lender shows up on its title.
Read vehicle titles from every state as layout-true text
LLMWhisperer turns a scanned or photographed car title into text that keeps each box and its label together, so the lien date never gets read as the issue date. It reads the odometer status boxes and the title brand boxes in their marked and unmarked states, straightens titles shot at an angle, and returns all 17 characters of the VIN exactly as printed.
That makes car title OCR a dependable first step: feed the text to your own LLM, or hand it to extraction.
Learn more about LLMWhispererTurn vehicle titles and lien releases into one lien record
List the fields your collateral record needs, and Document Studio returns them in the same JSON shape from a certificate of title, a title pledge form, a lien release or a registration application. This vehicle title extraction needs no state template: a new layout from a state you have never lent in comes back in the same shape. A VIN that differs from the retail installment contract, or a lienholder that is not you, comes back flagged with the reason.
Developers call it as an API from a dealer portal or a title loan app, back offices run it as an ETL pipeline over the titles inbox, or it becomes one step of a lien-perfection run in Flow.
Learn more about Document StudioConfirm your lien on every vehicle title before the file closes
Describe in plain English how your team checks a vehicle title against an auto loan or a title loan, and Flow's agents build the workflow. Each title is read on arrival, its VIN, owner and first lienholder are matched to the loan, and the lien position is recorded with the title image attached. You approve the workflow once, and every title runs through it the same way.
You decide where a person steps in: a title with no lien noted, a second lienholder ahead of you, a salvage or flood brand, or a VIN one character off from the contract. Only those titles stop, and a title clerk sees the title beside the loan with the difference marked.
Learn more about FlowRead your vehicle titles today. Automate lien checks when you're ready.
Start for free and run your own certificates of title through parsing and extraction now. Loan matching and recording the lien in your loan origination system start with a consultation.
Certificates of title and the paperwork that moves a lien
Documents
- Vehicle certificates of title
- Title pledge forms
- Lien releases
- Driver's Vehicle Registration Applications
- Applications for certificate of title
- Electronic title (ELT) printouts
- Title reassignment forms
- Odometer disclosure statements
- Powers of attorney for title transfer
- Duplicate title applications
- Bills of sale
- Salvage and rebuilt titles
- Motorcycle and RV titles
- Manufactured home titles
Fields captured
For the lenders whose collateral is parked in someone's driveway
Loan officers who take a paid-off vehicle as extra collateral see its title read, the VIN confirmed and any old lien flagged before the application goes to underwriting.
Processors stop keying VINs and lienholder names off dealer scans, and underwriters see the odometer, title brand and lien position in the file instead of a photocopy.
Smaller lending teams that handle titles by mail get each one read and matched to its loan on arrival, with the lien releases filed beside the paid-off accounts they close.
Indirect auto and title pledge desks follow every dealer funding until the state returns a title with their name on it, and the ones still missing a lien after 30 days surface on their own.
Liens confirmed while the loan is still new
Protecting the paper that proves who owns the car
A certificate of title ties an owner's name and home address to a VIN and to the lender holding the lien. A forged lien release on that paper is how a car gets sold out from under a lender, so every title and release is data worth guarding as closely as the loan itself.
- Titles and lien releases are encrypted in transit and at rest.
- Borrower titles are never used to train any model.
- Access by role and by lending desk, with every corrected VIN and cleared lien flag logged against the loan for your auditors.
- You set retention: title images are removed when your record policy says so, and the extracted lien record stays with the loan.
- Self-hosted deployment for lenders that keep collateral records inside their own environment.
What auto and title lenders ask about reading titles
The work splits in two. LLMWhisperer is the title pledge OCR step that reads the handwritten pledge form and the certificate of title behind it, and Document Studio deploys an API endpoint that returns the pledgor, VIN, odometer, loan amount and lienholder as JSON. POST the scan from your store or app and the record comes back in one call.
Unstract covers the whole pledge file: the title pledge form, the certificate of title and the lien release that ends the pledge. Document Studio extracts the fields with AI, and Flow's agents check each pledge against the loan and stop the ones where the title names a different owner or lender.
Each state lays out its title differently and names the lender in its own words. LLMWhisperer keeps every box with its label, and Document Studio reads the fields by meaning, so a title from a new state comes back in the same JSON shape with no template to build.
Yes. Document Studio deploys an endpoint per document type: send the certificate of title as a PDF or image and get the title number, VIN, vehicle, owner, odometer, lienholder and brand back as JSON. The same integration accepts lien releases and registration applications.
Yes. The first and second lienholders are extracted and compared with the lender on the loan. A title that shows no lien, another lender or a different VIN stops for a title clerk with the mismatch marked.
Yes. The odometer reading is extracted with its status box (actual, not actual or exceeds mechanical limits), and the brand section returns salvage, rebuilt, flood or none, so a branded vehicle is flagged before the loan is valued.
The release is read like the title: lienholder, release date and the authorized signature. It is matched to the paid-off loan, and a release that names a lender or VIN you do not hold is flagged instead of filed.
Yes. Driver's vehicle registration applications, applications for a certificate of title and ELT printouts from the state are read with the same approach, so a deal jacket with all three comes back as one set of records.
Yes. Borrowers and dealers send titles shot on a car seat or a desk, often folded. LLMWhisperer straightens the image and reads through creases and glare, and a VIN it cannot read in full comes back empty and flagged, never guessed.
Unstract is SOC 2, ISO 27001, GDPR and HIPAA compliant. Titles are encrypted in transit and at rest, never used to train models, and deleted on the schedule you set. Lenders that keep collateral data in-house can self-host; see the Trust Center.
Bring a week of dealer titles to the consultation
Titles from three states, a phone photo folded through the VIN, a title pledge form and the one that came back with no lien on it. We read them live and show the fields, the loan match and every flag.