--- a/C:\\xampp\\htdocs\\ALimpiezas\\ai-service\\main.py
+++ b/C:\\xampp\\htdocs\\ALimpiezas\\ai-service\\main.py
@@ -1,12 +1,15 @@
-\"\"\"Microservicio IA: OCR, voz e intenciones (ALimpiezas).\"\"\"
+\"\"\"Microservicio IA: OCR, comprensión fiscal ES y Ollama (ALimpiezas).\"\"\"
from __future__ import annotations
import io
+import json
import os
import re
import shutil
import tempfile
+import urllib.error
+import urllib.request
from datetime import date
from typing import Any
@@ -20,7 +23,7 @@
text: str = \"\"
-app = FastAPI(title=\"ALimpiezas AI Service\", version=\"0.2.0\")
+app = FastAPI(title=\"ALimpiezas AI Service\", version=\"0.3.0\")
app.add_middleware(
CORSMiddleware,
@@ -66,9 +69,37 @@
IMAGE_EXTENSIONS = (\".jpg\", \".jpeg\", \".png\", \".webp\", \".gif\", \".heic\", \".heif\", \".bmp\", \".tif\", \".tiff\")
+DEFAULT_OLLAMA_URL = os.getenv(\"OLLAMA_URL\", \"http://192.168.1.16:11434\")
+DEFAULT_OLLAMA_MODEL = os.getenv(\"OLLAMA_MODEL\", \"qwen2.5:7b\")
-def _parse_amount(value: str) -> float | None:
- cleaned = value.replace(\"€\", \"\").replace(\"EUR\", \"\").strip()
+FISCAL_JSON_SCHEMA = {
+ \"document_kind\": \"expense_received|issued_invoice|simplified_ticket|proforma|unknown\",
+ \"counterparty_name\": \"string\",
+ \"counterparty_cif\": \"string\",
+ \"counterparty_address\": \"string\",
+ \"invoice_number\": \"string\",
+ \"invoice_series\": \"string\",
+ \"invoice_date\": \"YYYY-MM-DD\",
+ \"operation_date\": \"YYYY-MM-DD or null\",
+ \"concept\": \"string\",
+ \"base_amount\": \"number\",
+ \"tax_rate\": \"21|10|4|0\",
+ \"tax_amount\": \"number\",
+ \"irpf_rate\": \"number or null\",
+ \"irpf_amount\": \"number or null\",
+ \"total_amount\": \"number\",
+ \"currency\": \"EUR\",
+ \"payment_method\": \"string\",
+ \"is_rectifying\": \"boolean\",
+ \"deductible\": \"boolean\",
+ \"notes\": \"string\",
+}
+
+
+def _parse_amount(value: str | None) -> float | None:
+ if not value:
+ return None
+ cleaned = str(value).replace(\"€\", \"\").replace(\"EUR\", \"\").replace(\" \", \"\").strip()
if cleaned.count(\",\") == 1 and cleaned.count(\".\") > 1:
cleaned = cleaned.replace(\".\", \"\")
cleaned = cleaned.replace(\",\", \".\")
@@ -78,6 +109,46 @@
return None
+def _normalize_date(day: str, month: str, year: str) -> str | None:
+ year_int = int(year)
+ if year_int < 100:
+ year_int += 2000
+ try:
+ return date(year_int, int(month), int(day)).isoformat()
+ except ValueError:
+ return None
+
+
+def _validate_spain_nif(value: str) -> bool:
+ value = re.sub(r\"[\\s.-]\", \"\", value.upper())
+ if not value:
+ return False
+
+ if re.fullmatch(r\"\\d{8}[A-HJ-NP-TV-Z]\", value):
+ letters = \"TRWAGMYFPDXBNJZSQVHLCKE\"
+ return value[8] == letters[int(value[:8]) % 23]
+
+ if re.fullmatch(r\"[XYZ]\\d{7}[A-HJ-NP-TV-Z]\", value):
+ prefix = {\"X\": \"0\", \"Y\": \"1\", \"Z\": \"2\"}[value[0]]
+ letters = \"TRWAGMYFPDXBNJZSQVHLCKE\"
+ num = prefix + value[1:8]
+ return value[8] == letters[int(num) % 23]
+
+ if re.fullmatch(r\"[A-HJ-NP-SUVW]\\d{7}[0-9A-J]\", value):
+ body, control = value[:8], value[8]
+ digits = sum(int(ch) for ch in body[1:] if ch.isdigit())
+ letter_map = \"JABCDEFGHI\"
+ if body[0].isdigit():
+ expected = str(digits % 10)
+ elif re.fullmatch(r\"\\d{7}\", body[1:]):
+ expected = letter_map[digits % 10]
+ else:
+ return False
+ return control == expected
+
+ return False
+
+
def _preprocess_image(image: Image.Image) -> Image.Image:
gray = image.convert(\"L\")
enhanced = ImageEnhance.Contrast(gray).enhance(2.0)
@@ -120,76 +191,332 @@
return \"\", \"none\"
+def _find_all_amounts(text: str) -> list[tuple[float, str]]:
+ found: list[tuple[float, str]] = []
+ patterns = [
+ r\"(?:TOTAL|T\\.?\\s*A\\.?\\s*PAGAR|IMPORTE\\s+TOTAL|SUMA\\s+TOTAL|AMOUNT\\s+DUE)\\s*[:.]?\\s*([\\d.,]+)\\s*€?\",
+ r\"(?:TOTAL\\s+FACTURA|TOTAL\\s+€)\\s*[:.]?\\s*([\\d.,]+)\",
+ r\"€\\s*([\\d.,]+)\\s*(?:$|\\n)\",
+ ]
+ upper = text.upper()
+ for pattern in patterns:
+ for match in re.finditer(pattern, upper, re.I | re.M):
+ amount = _parse_amount(match.group(1))
+ if amount is not None and amount > 0:
+ found.append((amount, match.group(0)[:80]))
+ return found
+
+
+def _classify_document_kind(text: str, doc_type_hint: str) -> str:
+ upper = text.upper()
+ if doc_type_hint == \"issued\":
+ return \"issued_invoice\"
+ if doc_type_hint == \"expense\":
+ if re.search(r\"\\bTICKET|SIMPLIFICAD|TPV|RECIBO\\b\", upper):
+ return \"simplified_ticket\"
+ if re.search(r\"\\bFACTURA\\s+RECIBIDA|PROVEEDOR|COMPRA\\b\", upper):
+ return \"expense_received\"
+ if re.search(r\"\\bFACTURA\\s+EMITIDA|CLIENTE|VENTA\\b\", upper):
+ return \"issued_invoice\"
+ return \"expense_received\"
+ if re.search(r\"\\bPROFORMA\\b\", upper):
+ return \"proforma\"
+ if re.search(r\"\\bFACTURA\\s+EMITIDA\\b\", upper):
+ return \"issued_invoice\"
+ if re.search(r\"\\bTICKET\\b\", upper):
+ return \"simplified_ticket\"
+ return \"expense_received\"
+
+
def _extract_from_text(text: str, doc_type: str) -> dict[str, Any]:
text = text.replace(\"\\x0c\", \"\\n\")
upper = text.upper()
- cif_match = re.search(r\"\\b([A-HJ-NP-SUVW]\\d{7}[0-9A-J]|\\d{8}[A-HJ-NP-TV-Z])\\b\", upper)
+ cif_match = re.search(
+ r\"\\b(?:CIF|NIF|N\\.?I\\.?F\\.?|DNI|C\\.?I\\.?F\\.?)\\s*[:.]?\\s*([A-HJ-NP-SUVW]\\d{7}[0-9A-J]|\\d{8}[A-HJ-NP-TV-Z]|[XYZ]\\d{7}[A-HJ-NP-TV-Z])\\b\",
+ upper,
+ )
+ if not cif_match:
+ cif_match = re.search(r\"\\b([A-HJ-NP-SUVW]\\d{7}[0-9A-J]|\\d{8}[A-HJ-NP-TV-Z]|[XYZ]\\d{7}[A-HJ-NP-TV-Z])\\b\", upper)
+
date_match = re.search(r\"\\b(\\d{1,2})[/.-](\\d{1,2})[/.-](\\d{2,4})\\b\", text)
- invoice_match = re.search(r\"(?:FACTURA|N[º°\\.]|NUM(?:ERO)?\\.?|FRA\\.?)\\s*[:#]?\\s*([A-Z0-9/-]+)\", upper)
- total_match = re.search(r\"(?:TOTAL|T\\.?\\s*A\\.?\\s*PAGAR|IMPORTE\\s+TOTAL)\\s*[:.]?\\s*([\\d.,]+)\", upper, re.I)
- base_match = re.search(r\"(?:BASE|BI|BASE\\s+IMPONIBLE)\\s*[:.]?\\s*([\\d.,]+)\", upper, re.I)
- iva_match = re.search(r\"(?:IVA|CUOTA\\s+IVA)\\s*[:.]?\\s*([\\d.,]+)\", upper, re.I)
+ invoice_match = re.search(
+ r\"(?:FACTURA|N[º°\\.]|NUM(?:ERO)?\\.?|FRA\\.?|Nº\\s*FACTURA)\\s*[:#]?\\s*([A-Z0-9/-]+)\",
+ upper,
+ )
- supplier_name = \"\"
- for line in [ln.strip() for ln in text.splitlines() if ln.strip()][:12]:
- if len(line) < 4 or re.search(r\"^\\d+[\\d\\s.,€]+$\", line):
- continue
- if re.search(r\"factura|ticket|cliente|fecha|total|iva|nif|cif\", line, re.I):
- continue
- supplier_name = line[:150]
- break
+ total_candidates = _find_all_amounts(text)
+ total_amount = total_candidates[-1][0] if total_candidates else None
- invoice_date = None
- if date_match:
- day, month, year = date_match.groups()
- year = int(year)
- if year < 100:
- year += 2000
- try:
- invoice_date = date(year, int(month), int(day)).isoformat()
- except ValueError:
- invoice_date = None
+ base_match = re.search(
+ r\"(?:BASE\\s+IMPONIBLE|BASE|BI|SUBTOTAL)\\s*[:.]?\\s*([\\d.,]+)\",
+ upper,
+ re.I,
+ )
+ iva_match = re.search(
+ r\"(?:CUOTA\\s+IVA|IVA\\s*(?:21|10|4)?%?|I\\.V\\.A\\.?)\\s*[:.]?\\s*([\\d.,]+)\",
+ upper,
+ re.I,
+ )
+ irpf_match = re.search(
+ r\"(?:IRPF|RETENCI[ÓO]N)\\s*(?:\\d+%)?\\s*[:.]?\\s*([\\d.,]+)\",
+ upper,
+ re.I,
+ )
- total_amount = _parse_amount(total_match.group(1)) if total_match else None
+ tax_rate = 21.0
+ rate_match = re.search(r\"IVA\\s*(21|10|4)\\s*%\", upper)
+ if rate_match:
+ tax_rate = float(rate_match.group(1))
+
base_amount = _parse_amount(base_match.group(1)) if base_match else None
tax_amount = _parse_amount(iva_match.group(1)) if iva_match else None
+ irpf_amount = _parse_amount(irpf_match.group(1)) if irpf_match else None
if base_amount is None and total_amount and tax_amount:
- base_amount = round(total_amount - tax_amount, 2)
+ base_amount = round(total_amount - tax_amount - (irpf_amount or 0), 2)
elif base_amount is None and total_amount:
- base_amount = round(total_amount / 1.21, 2)
- tax_amount = round(total_amount - base_amount, 2)
+ base_amount = round(total_amount / (1 + tax_rate / 100), 2)
+ tax_amount = round(total_amount - base_amount - (irpf_amount or 0), 2)
- confidence = 0.25
+ supplier_name = \"\"
+ for line in [ln.strip() for ln in text.splitlines() if ln.strip()][:15]:
+ if len(line) < 4 or re.search(r\"^\\d+[\\d\\s.,€]+$\", line):
+ continue
+ if re.search(r\"factura|ticket|cliente|fecha|total|iva|nif|cif|base|importe\", line, re.I):
+ continue
+ supplier_name = line[:150]
+ break
+
+ invoice_date = _normalize_date(*date_match.groups()) if date_match else None
+ counterparty_cif = cif_match.group(1) if cif_match else \"\"
+ document_kind = _classify_document_kind(text, doc_type)
+
+ confidence = 0.2
if len(text) > 50:
- confidence += 0.1
- if cif_match:
- confidence += 0.15
+ confidence += 0.08
+ if counterparty_cif and _validate_spain_nif(counterparty_cif):
+ confidence += 0.18
+ elif counterparty_cif:
+ confidence += 0.08
if invoice_date:
confidence += 0.1
if total_amount:
- confidence += 0.2
+ confidence += 0.22
if supplier_name:
confidence += 0.1
if base_amount:
+ confidence += 0.07
+ if invoice_match:
confidence += 0.05
return {
+ \"document_kind\": document_kind,
+ \"counterparty_name\": supplier_name,
+ \"counterparty_cif\": counterparty_cif,
+ \"counterparty_address\": \"\",
\"supplier_name\": supplier_name,
- \"supplier_cif\": cif_match.group(1) if cif_match else \"\",
+ \"supplier_cif\": counterparty_cif,
\"invoice_number\": invoice_match.group(1) if invoice_match else \"\",
+ \"invoice_series\": \"\",
\"invoice_date\": invoice_date or date.today().isoformat(),
+ \"operation_date\": invoice_date,
+ \"concept\": \"\",
\"base_amount\": base_amount,
+ \"tax_rate\": tax_rate,
\"tax_amount\": tax_amount,
+ \"irpf_rate\": None,
+ \"irpf_amount\": irpf_amount,
\"total_amount\": total_amount,
- \"confidence\": min(confidence, 0.98),
+ \"currency\": \"EUR\",
+ \"payment_method\": \"\",
+ \"is_rectifying\": bool(re.search(r\"\\bRECTIFICAT\", upper)),
+ \"deductible\": document_kind in (\"expense_received\", \"simplified_ticket\"),
+ \"confidence\": min(confidence, 0.92),
\"doc_type\": doc_type,
- \"notes\": \"Revisa los datos antes de confirmar.\",
- \"raw_text_preview\": text[:800],
+ \"notes\": \"\",
+ \"raw_text_preview\": text[:1200],
}
+def _ollama_extract(
+ text: str,
+ doc_type: str,
+ ollama_url: str,
+ model: str,
+) -> dict[str, Any] | None:
+ url = ollama_url.rstrip(\"/\") + \"/api/chat\"
+ kind_hint = \"factura recibida / gasto deducible\" if doc_type == \"expense\" else \"factura emitida / venta\"
+
+ system_prompt = (
+ \"Eres un experto fiscal español (RD 1619/2012, Ley IVA, autónomos). \"
+ \"Extrae datos del texto OCR de un documento contable español. \"
+ \"Responde SOLO con JSON válido, sin markdown. \"
+ \"Usa null para campos desconocidos. Importes en euros con 2 decimales. \"
+ \"document_kind: expense_received, issued_invoice, simplified_ticket, proforma o unknown. \"
+ \"Valida coherencia: base + IVA - IRPF ≈ total (±0.05€).\"
+ )
+ user_prompt = (
+ f\"Tipo esperado: {kind_hint} (doc_type={doc_type}).\\n\"
+ f\"Esquema JSON: {json.dumps(FISCAL_JSON_SCHEMA, ensure_ascii=False)}\\n\\n\"
+ f\"TEXTO OCR:\\n{text[:6000]}\"
+ )
+
+ payload = json.dumps({
+ \"model\": model,
+ \"messages\": [
+ {\"role\": \"system\", \"content\": system_prompt},
+ {\"role\": \"user\", \"content\": user_prompt},
+ ],
+ \"stream\": False,
+ \"format\": \"json\",
+ \"options\": {\"temperature\": 0.1},
+ }).encode(\"utf-8\")
+
+ req = urllib.request.Request(
+ url,
+ data=payload,
+ headers={\"Content-Type\": \"application/json\"},
+ method=\"POST\",
+ )
+
+ try:
+ with urllib.request.urlopen(req, timeout=90) as resp:
+ body = json.loads(resp.read().decode(\"utf-8\"))
+ except (urllib.error.URLError, TimeoutError, json.JSONDecodeError, KeyError):
+ return None
+
+ content = body.get(\"message\", {}).get(\"content\", \"\")
+ if not content:
+ return None
+
+ try:
+ parsed = json.loads(content)
+ except json.JSONDecodeError:
+ match = re.search(r\"\\{.*\\}\", content, re.S)
+ if not match:
+ return None
+ try:
+ parsed = json.loads(match.group(0))
+ except json.JSONDecodeError:
+ return None
+
+ if not isinstance(parsed, dict):
+ return None
+
+ return parsed
+
+
+def _merge_extractions(base: dict[str, Any], llm: dict[str, Any] | None) -> dict[str, Any]:
+ if not llm:
+ return base
+
+ merged = dict(base)
+ for key, value in llm.items():
+ if value is None or value == \"\":
+ continue
+ merged[key] = value
+
+ name = merged.get(\"counterparty_name\") or merged.get(\"supplier_name\") or \"\"
+ cif = merged.get(\"counterparty_cif\") or merged.get(\"supplier_cif\") or \"\"
+ merged[\"supplier_name\"] = name
+ merged[\"supplier_cif\"] = cif
+ merged[\"counterparty_name\"] = name
+ merged[\"counterparty_cif\"] = cif
+
+ if llm:
+ merged[\"confidence\"] = min(float(merged.get(\"confidence\", 0.5)) + 0.12, 0.98)
+ merged[\"ocr_engine\"] = (merged.get(\"ocr_engine\") or \"ocr\") + \"+ollama\"
+
+ return merged
+
+
+def _validate_fiscal_data(data: dict[str, Any]) -> dict[str, Any]:
+ errors: list[str] = []
+ warnings: list[str] = []
+ field_status: dict[str, str] = {}
+
+ required_expense = [\"counterparty_name\", \"invoice_date\", \"total_amount\"]
+ required_issued = [\"counterparty_name\", \"invoice_number\", \"invoice_date\", \"total_amount\"]
+
+ doc_kind = data.get(\"document_kind\", \"expense_received\")
+ is_issued = doc_kind == \"issued_invoice\" or data.get(\"doc_type\") == \"issued\"
+ required = required_issued if is_issued else required_expense
+
+ for field in required:
+ value = data.get(field) or data.get(field.replace(\"counterparty\", \"supplier\"))
+ field_status[field] = \"ok\" if value not in (None, \"\", 0) else \"missing\"
+ if field_status[field] == \"missing\":
+ label = {
+ \"counterparty_name\": \"Nombre emisor/receptor\",
+ \"invoice_number\": \"Número de factura\",
+ \"invoice_date\": \"Fecha\",
+ \"total_amount\": \"Importe total\",
+ }.get(field, field)
+ errors.append(f\"Falta dato obligatorio: {label}.\")
+
+ cif = data.get(\"counterparty_cif\") or data.get(\"supplier_cif\") or \"\"
+ if cif:
+ if _validate_spain_nif(cif):
+ field_status[\"counterparty_cif\"] = \"ok\"
+ else:
+ field_status[\"counterparty_cif\"] = \"review\"
+ warnings.append(\"El NIF/CIF no supera la validación algorítmica. Revísalo.\")
+ else:
+ field_status[\"counterparty_cif\"] = \"review\"
+ warnings.append(\"Sin NIF/CIF del emisor. Obligatorio en facturas completas (RD 1619/2012).\")
+
+ base = data.get(\"base_amount\")
+ tax = data.get(\"tax_amount\") or 0
+ irpf = data.get(\"irpf_amount\") or 0
+ total = data.get(\"total_amount\")
+
+ if base and total:
+ expected = round(float(base) + float(tax) - float(irpf), 2)
+ diff = abs(expected - float(total))
+ if diff <= 0.05:
+ field_status[\"total_amount\"] = \"ok\"
+ elif diff <= 0.5:
+ field_status[\"total_amount\"] = \"review\"
+ warnings.append(f\"Total ({total}€) no cuadra exactamente con base+IVA-IRPF ({expected}€).\")
+ else:
+ field_status[\"total_amount\"] = \"review\"
+ warnings.append(f\"Incoherencia fiscal: base ({base}€) + IVA ({tax}€) - IRPF ({irpf}€) ≠ total ({total}€).\")
+
+ if not total:
+ field_status[\"total_amount\"] = \"missing\"
+
+ tax_rate = data.get(\"tax_rate\")
+ if tax_rate not in (21, 10, 4, 0, 21.0, 10.0, 4.0, 0.0, None):
+ warnings.append(f\"Tipo de IVA inusual ({tax_rate}%). Verifica si aplica 21%, 10% o 4%.\")
+
+ if doc_kind == \"simplified_ticket\":
+ warnings.append(\"Ticket simplificado: conserva el original; algunos datos pueden no ser exigibles.\")
+
+ if data.get(\"is_rectifying\"):
+ warnings.append(\"Factura rectificativa detectada. Verifica referencia a la factura original.\")
+
+ compliance = 1.0
+ compliance -= 0.25 * sum(1 for s in field_status.values() if s == \"missing\")
+ compliance -= 0.1 * sum(1 for s in field_status.values() if s == \"review\")
+ compliance = max(0.0, min(compliance, 1.0))
+
+ data[\"validation_errors\"] = errors
+ data[\"validation_warnings\"] = warnings
+ data[\"field_status\"] = field_status
+ data[\"fiscal_compliance\"] = round(compliance, 2)
+ data[\"warnings\"] = warnings
+
+ if errors:
+ data[\"notes\"] = \"Revisa los campos marcados antes de confirmar. \" + (data.get(\"notes\") or \"\")
+ elif warnings:
+ data[\"notes\"] = \"Datos extraídos. Comprueba coherencia fiscal antes de confirmar.\"
+ else:
+ data[\"notes\"] = \"Datos fiscales coherentes. Confirma para registrar en contabilidad.\"
+
+ return data
+
+
def _detect_document_kind(filename: str, content_type: str) -> str:
lower = (filename or \"\").lower()
ct = (content_type or \"\").lower()
@@ -290,10 +617,12 @@
return {
\"status\": \"ok\",
\"service\": \"alimpiezas-ai\",
+ \"version\": \"0.3.0\",
\"tesseract\": HAS_TESSERACT,
\"pymupdf\": HAS_PYMUPDF,
\"heif\": HAS_HEIF,
\"whisper\": bool(os.getenv(\"OPENAI_API_KEY\")),
+ \"ollama_default\": DEFAULT_OLLAMA_URL,
}
@@ -301,6 +630,9 @@
async def extract_invoice(
file: UploadFile = File(...),
doc_type: str = Form(default=\"expense\"),
+ mode: str = Form(default=\"ocr\"),
+ ollama_url: str = Form(default=\"\"),
+ ollama_model: str = Form(default=\"\"),
) -> dict[str, Any]:
filename = file.filename or \"documento\"
content_type = file.content_type or \"\"
@@ -328,6 +660,19 @@
result = _extract_from_text(text, doc_type)
result[\"ocr_engine\"] = engine
+
+ if mode == \"ai\":
+ llm_url = (ollama_url or DEFAULT_OLLAMA_URL).strip()
+ llm_model = (ollama_model or DEFAULT_OLLAMA_MODEL).strip()
+ llm_data = _ollama_extract(text, doc_type, llm_url, llm_model)
+ result = _merge_extractions(result, llm_data)
+ if llm_data is None:
+ result.setdefault(\"validation_warnings\", []).append(
+ \"Ollama no respondió; se usaron solo reglas OCR. Comprueba la red LAN (192.168.1.16:11434).\"
+ )
+
+ result = _validate_fiscal_data(result)
+ result[\"source_file\"] = filename
return result