Intent Analysis and Data Catalog System

Conversational AI agent that analyzes user search intentions, detects ambiguities through intelligent clarification questions, and accesses structured dataset catalogs for data space applications.

---

Objectives

• Main objective: Demonstrate a robust intent analysis system for conversational data space agents that can extract structured search filters from natural language queries.

• Secondary goals:

  • Implement hybrid ambiguity detection (deterministic rules + LLM semantic analysis)
  • Prevent infinite clarification loops with adaptive questioning logic
  • Support multi-search sessions without context contamination
  • Provide extensible JSON-based dataset catalog architecture

• Expected impact: This research component enables more natural human-AI interaction in data space environments by intelligently handling ambiguous queries and maintaining conversation context across multiple searches.

---

Funding Information

This research project is supported by:

Funding organization/institution: MINISTERIO PARA LA TRANSFORMACION DIGITAL Y DE LA FUNCION PUBLICA Program or grant: CONVOCATORIA DE AYUDAS PROGRAMA DE ESPACIOS DE DATOS SECTORIALES PARA LA TRANSFORMACIÓN DIGITAL DE LOS SECTORES PRODUCTIVOS ESTRATÉGICOS MEDIANTE LA CREACIÓN DE DEMOSTRADORES Y CASOS DE USO DE ESPACIOS DE COMPARTICIÓN DE DATOS Project code/reference: TSI-100121-2024-24 Duration: [01/11/2024 – 31/12/2025]

---

Technology

This project uses the following technologies:

• Python 3.11+ - Programming language
• LangGraph - State machine framework for building conversational AI agents
• LangChain - LLM orchestration and prompt management
• Ollama - Local LLM server (Llama 3.1)
• uv - Fast Python package manager
• JSON - Dataset catalog storage format

Key Features

• ✅ Automatic extraction of structured filters (spatial, temporal, demographic)
• ✅ Hybrid ambiguity detection (deterministic + LLM)
• ✅ 3-attempt clarification limit to prevent infinite loops
• ✅ Multi-search session support with conversation boundaries
• ✅ Dynamic JSON-based dataset catalogs

For detailed architecture documentation, see docs.md.

Installation and Usage

Prerequisites

1. Python 3.11 or higher - Download

2. uv (fast Python package manager)

   • Windows PowerShell:

     powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

   • macOS/Linux:

     curl -LsSf https://astral.sh/uv/install.sh | sh

3. Ollama (local LLM server) - Download

   • After installation, run:

     ollama pull llama3.1

   • Verify it's running at http://127.0.0.1:11434

Installation Steps

1. Clone the repository

   git clone https://github.com/jjmb10-ua/entrega-clasificador.git
   cd entrega-clasificador

2. Install dependencies

   uv sync

3. Verify installation

   uv run python -c "import langgraph; print('✅ Installation successful')"

Running the Agent

Basic execution:

uv run python app.py

Example conversation:

>>> Tú: Busca datos de contaminación del aire en Madrid

Sistema: Para refinar la búsqueda, ¿podrías especificar?
         - Período temporal (año, rango de fechas)

>>> Tú: 2020-2024

Sistema: He entendido que buscas:
         - Tema: contaminación del aire
         - Ubicación: Madrid
         - Período: 2020-2024
         ¿Es correcto? (sí/no)

>>> Tú: sí

Exit: Type salir, exit, or quit

---

Authors / Contributors

• Juan José Martínez Berná – @jjmb10-ua

---

License

This project is distributed under the MIT License.

---

💬 Contact

For questions, collaborations, or further information:

📧 wake@dlsi.ua.es
🌐 Wake Research group

---

Documentation

For detailed technical documentation including:

• Complete architecture diagrams
• Execution flow examples
• Implementation details
• Design decisions
• Troubleshooting guide

See docs.md