The challenge
The platform helps law firms and lien-service companies manage construction-lien cases, from intake to sending statutory notices. Every new case started with someone reading a stack of documents (contracts, invoices, change orders, often scanned) and typing parties, addresses, amounts and work dates into the system. The client wanted AI to prepare that first draft, without ever letting it invent facts in a legal case.
What we built
Users upload up to 10 files (PDF, Word, Excel, images) or paste notes. Text is read directly from digital PDFs, and scans go through OCR. Rule-based extractors fill the fields they can; an AI model is called only for what the rules cannot settle, such as contract terms and unpaid work months, and every quote it returns is checked against the source text. The result is a draft case: a review screen shows each value with its confidence, source page and quote, and nothing goes live until a person confirms it. Deadlines are calculated by the application, never by the AI.
Start from a sentence or a contract
The dashboard asks what you are working on. Type the job in plain words, or drop in the contract, invoices and change orders.
Open this screen full size ↗Real screens from the product. Client branding and every case detail are replaced with sample data.
How we worked
We worked on both the extraction engine and the review screens. The original extraction ran as a separate Python service; behind a stable interface we replaced it with an in-process Java engine, without changing the screens users already knew. We also set up the platform’s AWS deployment with HTTPS across multiple domains.