AI Data Acquisition Overview

AI data acquisition is the current architecture for converting uploaded rental-management documents into reviewed, validated business data.

The system must not treat AI output as final database truth. AI is used to accelerate data entry, pre-fill forms, and suggest structured records. The user and backend validation remain the control points before persistence.

Main flow

User upload
  -> document storage
  -> OCR if needed
  -> AI extraction API
  -> structured JSON
  -> confidence scoring
  -> pre-filled UI
  -> user review
  -> backend validation
  -> database save

Typical use case

1. A user uploads a document from a form, such as a rental contract or utility bill.
2. The backend stores the original file and creates a processing job.
3. OCR is executed only when the document does not already contain reliable text.
4. An AI model classifies the document and extracts structured data.
5. The AI returns JSON with field-level confidence.
6. The UI pre-fills the relevant form.
7. The user reviews, corrects, and confirms.
8. The backend validates the payload against business rules.
9. Validated data is saved to the application entities.

Current boundaries

• AI may suggest parties, properties, contracts, payments, bank_accounts, expected_expenses, and document metadata.
• AI must not create, update, or delete database records directly.
• Every AI-generated field must be traceable to a document, extraction timestamp, model, and confidence score.
• Low-confidence or ambiguous fields must be highlighted in the UI.

Primary design decisions

• Keep raw documents, extracted text, AI JSON, and final database records separate.
• Prefer deterministic backend validation after AI extraction.
• Store enough extraction metadata to debug errors and improve prompts later.
• Design the pipeline as asynchronous because OCR and model calls may be slow.
• Support retries and fallback models without changing the business data model.