Contract data extraction
Extract parties, subject, amounts, Jalali dates, term, obligations, and deadlines, with the source clause and page shown next to each item.
Legal AI for Persian contract and document review
An assistant for legal teams that reads Persian contracts, extracts clauses, parties, amounts, obligations, and deadlines, and ties every finding to the exact clause. Legal judgment and final sign-off stay with your lawyer.
An assistant for legal teams, not an AI lawyer
If you are looking for an "AI lawyer" that gives an opinion in place of a lawyer, this service is not that. It does not give legal advice, issue final legal opinions, or replace counsel. Its job is to make reading, extraction, and comparison easier for the legal team and to put every finding in front of the lawyer with a reference to the contract text. Interpretation and any conclusion under Iranian law stay with qualified counsel.
Common use cases
Comparison with your templates
Compare a contract with your approved templates and review checklist, and flag clauses that are missing, changed, or out of line with your position.
Contract archive search
Find similar clauses across past contracts, for example every contract with an arbitration clause or a late-delivery penalty.
Obligations and deadlines
Pull payment, delivery, renewal, and termination deadlines into one list the legal team can review or move into its existing calendar or work queue.
Scanned contracts
Turn paper contracts and scanned files into searchable text with Persian OCR, keeping the original image alongside the text.
First-pass summary
Draft an initial summary and a list of points to check, for the lawyer to correct and approve.
How we work
The assistant is measured on your own contracts, not on a prepared demo, and your legal team defines what counts as correct.
- 01
One contract type
With the legal team we pick one frequent contract type, such as purchase or construction contracts, and document the data and points to extract.
- 02
Reference templates and checklist
Approved templates, acceptable fallback wording, and unacceptable clauses come from your legal team, so the comparison follows your organisation's position rather than a generic model.
- 03
Reading and extraction
Scanned contracts go through Persian OCR; Jalali dates, amounts, and digits are normalised; and each finding is stored with the quoted clause text and page number.
- 04
Evaluation on past contracts
On past contracts your lawyers have already reviewed, we measure what the assistant found correctly, what it missed, and where it was wrong.
- 05
Deployment and monitoring
In daily use we keep the corrections the lawyer makes to outputs and use them in later evaluations.
Controls needed from day one
- Lawyer sign-off
- No output becomes the basis of a legal decision before a lawyer approves it. The assistant suggests findings; the legal team decides.
- Citation to clause
- Every finding points to the quoted clause text, clause number, and page. When the assistant cannot find something in the text, it says "not found" or "unclear" instead of guessing.
- Confidentiality
- The assistant can run on your own infrastructure so contracts stay inside the organisation. Your contracts are not used to train models.
- Access by matter
- Each user sees only the contracts and matters assigned to them, and the assistant builds answers only from those documents.
- Audit log
- The original document, the assistant's findings, the lawyer's edits, and the final approval are recorded with date and user so every decision can be traced.
Zharf
Try the ZharfAI assistant right now
Ask about Persian assistants, document automation, or how to start a pilot.
Frequently asked questions about legal AI
Is this an AI lawyer?
No. It is an assistant for the legal team: it reads contracts, finds clauses and deadlines, and shows where a contract differs from your templates. It does not give legal advice or issue final opinions, and conclusions under Iranian law stay with qualified counsel. A language model can be confidently wrong or miss a point, which is why every finding cites its source clause and a lawyer approves it.
How is accuracy checked?
On your own past contracts that your lawyers have already reviewed. The assistant's findings are compared with the lawyer's review, and three measures are reported separately: missed findings, wrong findings, and citation correctness. No accuracy number is promised before that evaluation.
Do our contracts leave the organisation?
Not necessarily. The assistant can run on your own infrastructure so contract text stays inside the organisation, or be hosted inside Iran. Data flow, storage location, retention, and per-user access to matters are documented before the project starts.
Which documents does it work with?
Persian contracts and amendments, whether as text files or scanned PDFs. Scan quality, stamps, and layout affect results. Handwriting, such as margin notes and handwritten corrections, has a higher error rate and usually needs a person to read or confirm it.
How much does legal AI cost?
Cost depends on contract types and volume, scan quality, number of users, deployment location, and integration with current systems. After discovery you receive a staged estimate: first for a pilot on one contract type, then for expansion to other contracts.
Start with one contract type
In the first conversation we review which contract type takes most of the legal team's time, which templates and checklist you have, and where contracts are stored. If a pilot is not a good fit, we say so plainly.