PerfectLien/Construction · LegalAI
AI document intake for construction lien software.
PerfectLien is a case management SaaS for US law firms and lien service companies: intake, preliminary lien notices, lien affidavits, deadlines, billing and a client portal. It also reads the paperwork. Upload invoices, contracts or scanned notices and the system drafts the case for a person to confirm.
Fig. 01 · PerfectLien · Lien case management with AI document intake · line drawing, not a screenshot
- measured AI cost per 3 page document
- $0.0064measured AI cost per 3 page document
- AI features shipped in the product
- 6AI features shipped in the product
- interface languages
- 4interface languages
Miss a lien deadline by one day and the right to file is gone.
Construction lien work runs on statutory deadlines. Each case starts as a pile of documents (invoices, contracts, change orders, scanned notices) that someone has to read and key in before the clock can be managed.
Manual intake is slow and error prone, and a wrong date or a missing party can cost a client their right to file. Any AI in this process has to be accurate, explain where every value came from, and never be the one to decide a legal deadline.
How the AI works.
- 01Read every documentPDFs with a text layer are read directly; scans go through Tesseract OCR at 300 DPI with an orientation retry. Word, Excel, images and pasted notes are accepted too, up to 10 files per upload.
- 02Rules first, the model secondA rule based extractor fills the fields it can find with confidence. Only if required fields are still missing does one batched LLM call handle the legal reasoning fields. When the rules are enough, the model is skipped entirely.
- 03Show the source of every valueThe review screen shows each field with its confidence, the source page and the verbatim quote it came from. A person confirms before the case goes live; uploads that match an existing case become a field by field merge proposal.
- 04Summaries, Q&A and an agentThe AI also summarises every document on a case, answers questions grounded in the case file, and a tool calling case agent proposes updates that a person applies. Nothing is saved without confirmation.
AI you can check.
- 01
Every quote the model returns is rechecked against the source text; an unverified value gets zero confidence.
- 02
Deadlines are always calculated in code, never by the model.
- 03
The AI proposes, a person applies: extracted cases start as drafts and agent changes need confirmation.
- 04
Cost controls: the LLM is skipped when rules suffice, documents per prompt are capped, and API keys rotate on rate limits.
What our team does
We revamped the AI behind document intake and case summaries, accuracy as well as the interface, and set up the AWS deployment with HTTPS across multiple domains.
Now · Live on AWS, served on multiple domains.
- Built
- AI document intake
- AI case summary & Q&A
- AWS deployment
- Stack
- Java 21
- Spring Boot
- React 19
- Tesseract OCR
- LLM + embeddings
- AWS
Questions we get about it.
How does AI document intake work for construction lien cases?
Uploaded documents are read (with OCR for scans), a rule based extractor fills the fields it can, and an LLM handles only the fields the rules can't. Every value is shown with its source page and quote, and a person confirms the draft case before it goes live.
How do you stop the AI from making things up?
Every quote the model returns is checked against the original document; anything that can't be verified is dropped. Legal deadlines are calculated in code, never by the model, and nothing is saved to a case without a person confirming it.
What does the AI cost to run per document?
We measured about $0.0064 for a three page document. When the rules find every required field, the language model isn't called at all.
Can you add AI document intake to our own software?
Yes. The same approach (OCR and rules first, an LLM only where needed, quotes verified, a person confirming) can be built into other document heavy workflows. Tell us what you receive and what you key in today.
Want AI like this in your software?
We’ll come back within one business day with how it could work, who would build it and what it would take.