A SaaS platform for construction companies in Santiago, Chile, needed to gather the information from each contract and, at the same time, build a shared base for looking up projects, areas and background information. The initial approach meant entering every piece of data by hand after uploading the document, just as the repository had to get ready for more than 5,000 contracts.
← All projectsSantiago de Chile · LLM · Contract management
AI-powered SaaS · Construction in Chile
One contract. The whole project, one question away.
How we turned a contract PDF into a structured, reviewable record you can query inside the product itself.
planned repository+5,000
01+5,000Planned contracts
021 PDFStarts the record
032Access levels
The core idea
A contract stops being an isolated PDF and becomes a structured record that can be reviewed and queried with artificial intelligence.
01 · The starting point
The contract held the truth. The product made you type it again.
Uploading thousands of documents solved nothing if every company, date, amount, person in charge and material then had to be entered by hand.
Duplication
The document and the form asked for the same information twice.
Structure
Each contract could include different annexes and org charts.
Privacy
Querying shared knowledge must not expose private data.
02 · The system
From a document blueprint to a knowledge base with permissions.
The AI proposes the structure, a person validates the fields, and the LLM answers according to each user's relationship with the contract.
Providencia
READING
AIextracts
CONTRACT OWNERSees private and financial dataFULL ACCESS
REST OF THE TEAMQueries general knowledgeGLOBAL VIEW
Who is in charge of this project and what materials are planned?
✦Answer built from the authorized contract.
- 01Interpret
Contract and annexes
The system reads the document and identifies its real structure.
- 02Pre-fill
Fields and org chart
Amounts, areas, dates, materials and people in charge come pre-filled.
- 03Validate
Human correction
A person reviews only what the AI may have misread.
- 04Query
Permission-aware LLM
Each question gets the level of detail that user is allowed to see.
Solution in place
We designed a flow that reads the contract and its annexes and pre-fills the company, address, amounts, purpose, materials, dates, floor areas, people in charge and org chart. A person reviews and corrects only the fields that need it before saving. Inside the product, an LLM lets the owner query their entire contract and gives other users general answers about areas, construction work, materials and budgets without exposing private information.
03 · The result
Data entry became a review, not a transcription.
The system already structures contracts inside the product and lets users query projects, areas and past records without losing access control.
Planned contracts
Starts the record
Access levels
What changed
Document upload is now built into the product, and the information is structured from the very first moment, with human review before it's consolidated. The system already answers questions with different permission levels, and the architecture is ready to take in a planned repository of more than 5,000 contracts; that figure is the target volume, not documents already processed.
“A document adds more value when it stops being a file and starts answering.”