Manually copying data from technical datasheets is slow and error-prone. [Adjust with the real case.]
AI & AUTOMATION · CASE 02
AI-powered PDF data extraction
Turn technical PDF datasheets into structured data and measure extraction accuracy.
An LLM with a strict JSON schema extracts fields; low-confidence fields are sent for human review.
[Field-level accuracy measured against a manually reviewed dataset.]
Extraction accuracy by field (example)preview · synthetic data · Power BI embed goes here
extraccion-pdf-ia/
├─ src/
│ ├─ leer_pdf.py
│ ├─ extraer.py
│ └─ esquema.py
├─ eval/
│ └─ set_revisado.csv
└─ README.md
class Ficha(BaseModel):
modelo: str
potencia_kw: float | None
voltaje: str | None
ficha = llm.extraer(texto, esquema=Ficha)
if ficha.confianza < 0.8:
enviar_a_revision(ficha)