Data & systems01

RDO Intelligence

A pipeline that turns hundreds of construction-site PDFs into an auditable database people can query in natural language.

PDFs consolidated
less preparation time
natural-language access
Document ingestionPDF extractionData integrity
PythonFastAPIPostgreSQLLlamaParse
Também usado
Private code - demonstration available on request

At a glance

ProblemCritical site facts were trapped across hundreds of PDFs.
SolutionA reliable ingestion pipeline and natural-language access through MCP.
Result200+ daily reports consolidated and 90% less preparation time, with traceable evidence.

The problem

For a multinational port construction contract, proving third-party delays meant searching months of daily reports by hand. The evidence was scattered across hundreds of PDF documents and spreadsheets.

That manual process was slow, hard to audit, and vulnerable to a missed fact at exactly the moment a contractual claim needed defensible evidence.

Impact

Around 200 reports became a structured, auditable source of truth.

Natural-language questions cut preparation time by 90% while preserving a path back to the source record.

The result supports more rigorous contractual analysis without replacing human judgment.

An auditable source of truth reduces search time and omission risk through the same change.

Architecture

The pipeline moves from raw PDF to reliable, queryable evidence.

Diagrama da arquitetura do RDO Intelligence
Fig. 01 — Arquitetura do pipeline
IngestionGoogle Drive directory monitoring
Orchestrationn8n in Docker on a VPS
ParsingLlamaParse for workforce, weather, and daily notes
APIFastAPI validation layer
DatabaseStructured and indexed PostgreSQL
InterfaceRead-only MCP connector