Challenge
A law firm drafts numerous purchase contracts for residential development projects. These are based on purchase offers, utility value appraisals (Nutzwertgutachten) and other supporting documents, which come in different content and file formats. The finished purchase contract, by contrast, always has the same structure. This workflow was to be automated.
The solution had to be able to process entire properties in a single pass. Because personal data such as bank details and ID numbers are processed, it had to be GDPR-compliant. At the same time, the accuracy of language models alone is not sufficient for a legally binding contract. Every extraction therefore needs to be checked by experts.
Solution
For this purpose, hpc.one designed and developed a web application for the entire purchase process. Utility value appraisals are read out on a rule basis and create the inventory of all flats, parking spaces and ancillary rooms. Purchase offers can be uploaded in batches. A language model extracts buyers, prices and payment schedules from them, which the application automatically assigns to the matching units.
In a review view, users check and correct the data before it is accepted. Purchase contracts and supporting documents are then generated as DOCX and PDF from stored templates with placeholders, also for several flats at once. The language model from a European provider runs in a dedicated server environment, sensitive fields are stored in encrypted form, and an audit log makes changes traceable.
Topics
- AI & Data
- Software Development
- Document Automation
- Process Management
- Data Protection and GDPR
- Quality Management
Methods
- Large Language Models
- OCR
- Python FastAPI
- Next.js
- PostgreSQL
- Docker
AI-Assisted Drafting of Purchase Contracts