Skip to main content

​
Overview

The /siaa/enrutar endpoint simulates the document routing process without actually extracting fragments or calling the AI model. It shows which documents would be selected for a given query and why, making it essential for debugging retrieval issues.

​
Endpoint

GET /siaa/enrutar?q=<pregunta>

​
Parameters

​
q
string
required
The query/question to test against the routing algorithmExample: ¿Cuál es la periodicidad del reporte SIERJU?

​
Response

​
pregunta
string
The original query string
​
doc_especifico
boolean
Whether the query contains specific document references (e.g., “PSAA16”, “Acuerdo”, “Circular”). When true, only 1 document is returned; otherwise, up to MAX_DOCS_CONTEXTO documents are returned.
​
max_docs_usados
integer
Maximum number of documents that would be used for this query (1 for specific doc queries, 2 for general queries in the default configuration)
​
docs_encontrados
array of objects
List of documents selected by the routing algorithm, in ranked order

​
Example

​
Request

curl "http://localhost:5000/siaa/enrutar?q=¿Cuál+es+la+periodicidad+del+reporte+SIERJU?"

​
Response

{
  "pregunta": "¿Cuál es la periodicidad del reporte SIERJU?",
  "doc_especifico": false,
  "max_docs_usados": 2,
  "docs_encontrados": [
    {
      "doc": "acuerdo_no._psaa16-10476.md",
      "tamano": 45826,
      "coleccion": "general",
      "chunks": 38
    },
    {
      "doc": "guia_sierju_despachos.md",
      "tamano": 12453,
      "coleccion": "general",
      "chunks": 15
    }
  ]
}

​
Request with Specific Document Reference

curl "http://localhost:5000/siaa/enrutar?q=¿Qué+dice+el+artículo+5+del+PSAA16?"

​
Response

{
  "pregunta": "¿Qué dice el artículo 5 del PSAA16?",
  "doc_especifico": true,
  "max_docs_usados": 1,
  "docs_encontrados": [
    {
      "doc": "acuerdo_no._psaa16-10476.md",
      "tamano": 45826,
      "coleccion": "general",
      "chunks": 38
    }
  ]
}

​
Use Cases

​
Debugging Wrong Document Selection

When users report that the system is returning answers from the wrong document: 
curl "http://localhost:5000/siaa/enrutar?q=¿Quién+debe+capacitar+sobre+SIERJU?"
Verify that the expected document appears in docs_encontrados. If not, check:
  • Keywords for the expected document (/siaa/keywords/<doc>)
  • Term density for query terms (/siaa/densidad/<term>)
  • Manual keyword configuration in KEYWORDS_MANUALES

​
Testing Query Variations

Compare how different phrasings affect document selection: 
# Formal phrasing
curl "http://localhost:5000/siaa/enrutar?q=periodicidad+del+formulario+SIERJU"

# Casual phrasing
curl "http://localhost:5000/siaa/enrutar?q=cada+cuánto+reporto+en+SIERJU"

​
Validating Multi-Level Routing

The routing algorithm uses three levels:
  1. TF-IDF keywords (automatic + manual) — weight 2.0
  2. Term density across documents — weight 1.0
  3. Filename token matching — weight 1.5
Test whether specific document names are being matched: 
curl "http://localhost:5000/siaa/enrutar?q=acuerdo+psaa16+10476"
This should strongly favor the matching document due to filename token scoring.

​
Verifying Specific Document Detection

The system limits results to 1 document when queries mention specific documents: 
# Should set doc_especifico = true
curl "http://localhost:5000/siaa/enrutar?q=circular+2023"

# Should set doc_especifico = false
curl "http://localhost:5000/siaa/enrutar?q=periodicidad+de+reportes"

​
Notes

  • The routing algorithm uses a multi-level scoring system combining TF-IDF, term density, and filename matching
  • Specific document patterns include: psaa, pcsja, acuerdo, circular, resolución, decreto
  • The MAX_DOCS_CONTEXTO configuration (default: 2) determines the maximum documents for general queries
  • Documents are scored but this endpoint doesn’t show the actual scores (check server logs with [ENRUTADOR] prefix for score details)
  • Query expansion happens automatically for temporal queries (“cuándo” → adds “periodicidad”, “plazo”, “hábil”)

Build docs developers (and LLMs) love

Get started for freeTalk to us