NLP Intent and Slot Recognition
General Description
This component provides natural language processing capabilities for intent detection and slot filling using BERT-based models. It processes text input to identify user intentions and extract structured information from the text, such as entities and their values. The system uses a fine-tuned BERT model with custom classifier heads for intent recognition and slot identification, served through a FastAPI interface. Training is performed on annotated data with Lightning, allowing for efficient model iteration and deployment in containerized environments.
| Resource | Link |
|---|---|
| Source code | https://gitc.piap.lukasiewicz.gov.pl/ai-prism/wp4/ai-based-perception-modules/nlp-intent-and-slot |
| Demo Video |
Contact
The following table includes contact information of the main developers in charge of the component:
| Name | Organisation | |
|---|---|---|
| Dorin Clisu | dorin.clisu@nttdata.com |  |
License
Proprietary.
Technical Foundations
Integrated and Open Source Components
Overview
The NLP Intent and Slot Recognition component integrates several key open-source libraries to provide its functionality. It leverages Hugging Face's Transformers for pre-trained language models, PyTorch Lightning for training workflows, FastAPI for serving predictions, and Uvicorn for ASGI server implementation. These components work together to enable efficient natural language understanding in robotics applications.
Pre-existing Components
FastAPI
Source
FastAPI is a modern web framework for building APIs with Python. https://fastapi.tiangolo.com/
Description
FastAPI is a high-performance web framework for building APIs with Python 3.7+ based on standard Python type hints. It provides automatic Swagger UI documentation, data validation, and serialization.
Modifications
None.
Purpose in AI-PRISM
FastAPI is used to expose the NLP model's capabilities via a RESTful API, allowing other AI-PRISM components to make intent and slot recognition requests and receive structured JSON responses.
License
MIT License https://github.com/tiangolo/fastapi/blob/master/LICENSE
Lightning (PyTorch Lightning)
Source
PyTorch Lightning is a lightweight PyTorch wrapper for high-performance AI research. https://lightning.ai/
Description
Lightning is a framework that organizes PyTorch code to separate research code from engineering code, making deep learning projects more maintainable and scalable.
Modifications
None.
Purpose in AI-PRISM
Lightning is used to structure the training workflow for the BERT-based model, handling the training loop, validation, checkpointing, and logging in a standardized way.
License
Apache License 2.0 https://github.com/Lightning-AI/lightning/blob/master/LICENSE
PyTorch
Source
PyTorch is an open-source machine learning framework. https://pytorch.org/
Description
PyTorch is a deep learning framework that provides a seamless path from research prototyping to production deployment with GPU acceleration.
Modifications
None.
Purpose in AI-PRISM
PyTorch is the core machine learning framework used to define, train, and deploy the neural network models for intent recognition and slot filling.
License
BSD-3-Clause License https://github.com/pytorch/pytorch/blob/master/LICENSE
Transformers
Source
Transformers is a library of state-of-the-art pre-trained models for natural language processing. https://huggingface.co/transformers/
Description
Transformers provides thousands of pre-trained models for various NLP tasks, including BERT models used in this component for contextual word embeddings.
Modifications
None.
Purpose in AI-PRISM
Transformers is used to load pre-trained BERT models that provide powerful contextual embeddings for text understanding, which are then fine-tuned for the specific intent and slot recognition tasks.
License
Apache License 2.0 https://github.com/huggingface/transformers/blob/main/LICENSE
Uvicorn
Source
Uvicorn is an ASGI web server implementation for Python. https://www.uvicorn.org/
Description
Uvicorn is a lightning-fast ASGI server implementation, based on uvloop and httptools.
Modifications
None.
Purpose in AI-PRISM
Uvicorn serves as the production ASGI server that hosts the FastAPI application, handling HTTP requests and responses efficiently.
License
BSD-3-Clause License https://github.com/encode/uvicorn/blob/master/LICENSE.md
How to install
Every AI-PRISM component is installed using the Cluster management service. During the installation process, the user needs to configure a set of high-level parameters.
How to use
The NLP Intent and Slot Recognition component can be used by sending HTTP POST requests to the /recognize endpoint with text to be analyzed. The service returns structured JSON with detected intent and extracted slots.