Datasetpublic

Ionutcroitoru/pbd-autism-caregiver

Privacy-by-Design in AI-Assisted Systems for Caregivers of Children with Autism: A Secure Multi-Agent Architecture Dataset Description This dataset accompanies the paper "Privacy-by-Design in AI-Assisted Systems for Caregivers of Children with Autism: A Secure Multi-Agent Architecture". The system is a privacy-by-design multi-agent architecture integrating Retrieval-Augmented Generation (RAG), Data Loss Prevention (DLP), consent management, explainable AI (XAI)… See the full description on the dataset page: https://huggingface.co/datasets/Ionutcroitoru/pbd-autism-caregiver.

sourceHugging Facecc-by-4.0updated 7mo agoView on Hugging Face

1likes23downloads

Dataset Card

Privacy-by-Design in AI-Assisted Systems for Caregivers of Children with Autism: A Secure Multi-Agent Architecture

Dataset Description

This dataset accompanies the paper "Privacy-by-Design in AI-Assisted Systems for Caregivers of Children with Autism: A Secure Multi-Agent Architecture". The system is a privacy-by-design multi-agent architecture integrating Retrieval-Augmented Generation (RAG), Data Loss Prevention (DLP), consent management, explainable AI (XAI), and audit trail capabilities to support caregivers of children with Autism Spectrum Disorder (ASD).

Dataset Summary

Config	Records	Description
`caregiver_qa`	150	Question-answer pairs from synthetic caregiver observation notes
`caregiver_notes`	200	Synthetic caregiver observation notes about a child with ASD
`dsm5_qa`	100	Medical knowledge QA pairs based on DSM-5 diagnostic criteria

Loading the Dataset

python

from datasets import load_dataset

# Load a specific configuration
qa_dataset = load_dataset("YOUR_USERNAME/pbd-autism-caregiver", "caregiver_qa")
notes_dataset = load_dataset("YOUR_USERNAME/pbd-autism-caregiver", "caregiver_notes")
dsm5_dataset = load_dataset("YOUR_USERNAME/pbd-autism-caregiver", "dsm5_qa")

Dataset Configurations

`caregiver_qa` (150 records)

Question-answer pairs generated from synthetic caregiver observation notes across 5 behavioral categories.

Field	Type	Description
`id`	`string`	Unique identifier (e.g., `mihaela_qa_0000`)
`question`	`string`	Natural language question about the child's behavior
`gold_answer`	`string`	Ground truth answer
`context`	`string`	Source caregiver note text
`category`	`string`	`transitions` \	`communication` \	`sensory` \	`social` \	`behavior`
`difficulty`	`string`	`easy` \	`medium` \	`hard`
`hash`	`string`	Content hash for deduplication
`source`	`string`	Data source identifier
`note_id`	`string`	Reference to source note

`caregiver_notes` (200 records)

Synthetic caregiver observation notes simulating real-world autism care documentation. Each note includes explicitly marked PII elements for DLP evaluation.

Field	Type	Description
`note_id`	`string`	Unique note identifier (UUID)
`child_id`	`int`	Synthetic child identifier
`child_name`	`string`	Synthetic child name
`caregiver_id`	`int`	Synthetic caregiver identifier
`specialist_id`	`int`	Synthetic specialist identifier
`text`	`string`	Full observation note text
`behavior_context`	`string`	Observation context (e.g., "Home - morning routine")
`category`	`string`	Behavioral category
`pii_elements`	`list`	List of PII elements with `type` and `value`

`dsm5_qa` (100 records)

Medical knowledge QA pairs based on DSM-5 diagnostic criteria for Autism Spectrum Disorder.

Field	Type	Description
`id`	`string`	Unique identifier (e.g., `dsm5_qa_0000`)
`question`	`string`	Medical question about ASD diagnosis/management
`gold_answer`	`string`	Ground truth answer based on DSM-5
`context`	`string`	Source context (empty for static knowledge)
`category`	`string`	`medical` \	`diagnostic` \	`behavioral` \	etc.
`difficulty`	`string`	Question difficulty level
`source`	`string`	`dsm5_static`

System Configuration

—LLM: Meta Llama 3 8B Instruct (local, via LM Studio)
—Embedding: mxbai-embed-large-v1 (1024 dimensions)
—Vector DB: Qdrant
—PII Detection: Microsoft Presidio + custom Romanian recognizers
—Infrastructure: Docker Compose (Kafka, NATS, Redis, Qdrant)

Synthetic Data Statement

All personal data in this dataset is entirely synthetic. Names, dates, and behavioral descriptions were generated programmatically and do not correspond to real individuals. The pii_elements field in caregiver_notes explicitly marks all synthetic PII for transparency and DLP evaluation purposes.

Citation

bibtex

@dataset{croitoru2026pbd_autism,
  author    = {Croitoru, Ionut and Turcu, Cristina and Turcu, Cornel},
  title     = {Dataset for: Privacy-by-Design in AI-Assisted Systems for Caregivers of Children with Autism: A Secure Multi-Agent Architecture},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Ionutcroitoru/pbd-autism-caregiver}
}

Authors

—Ionut Croitoru - Stefan cel Mare University of Suceava, Romania (ionut.croitoru@usm.ro)
—Cristina Turcu - Stefan cel Mare University of Suceava, Romania (cristina.turcu@usm.ro)
—Cornel Turcu - Stefan cel Mare University of Suceava, Romania (cornel.turcu@usm.ro)