Empathic Machine Processors for Inclusion (EMPI) is a project that supports the inclusion of people with special needs.
EMPI is a set of Conversational AI and mobile development tools for inclusive education, mental well-being and harmonious social interaction. Visit EMPI web-page to learn more, and contact Victoria Firsanova to contribute.
!git clone https://github.com/vifirsanova/empi.git
import sys
import os
sys.path.append(os.path.abspath('/content/empi/modules/'))
You can import all the necessary modules from EMPI modules or use custom compatible modules.
- AI-driven Teaching Assitant for inclusive education (uses GPU)
- EMPI AI Chat-bot greetings scenario
EMPI Dataset
The EMPI Dataset by Victoria Firsanova is a closed graph knowledge base. The EMPI Dataset enables personal data cyphering, and retrieval-augmented generation. Feel free to learn more or contribute.
The ASD QA Dataset
The ASD QA dataset by Victoria Firsanova is used to train Word Embeddings for the model.
The detailed dataset statistics
Parameter | Description |
---|---|
The number of QA pairs | 4,138 |
The number of irrelevant questions | 352 |
The average question length | 53 symbols / 8 words |
The average answer length | 141 symbols / 20 words |
The average reading paragraph length | 453 symbols / 63 words |
Max question length | 226 symbols / 32 words |
Max answer length | 555 symbols / 85 words |
Max reading paragraph length | 551 symbols / 94 words |
Min question length | 9 symbols / 2 words |
Min answer length | 5 symbols / 1 words |
Min reading paragraph length | 144 symbols / 17 words |
The ASD QA dataset is available on HuggingFace:
The model combines the power of Blockchain with Conversational AI technologies to create personal user blocks that store user data, app settings and chat logs for interpretability research.
{
"username": "Аня",
"init_log": "ЭМПИК:\nПривет! Как тебя зовут?\nПриветик! Меня зовут Аня.\nЭМПИК:\nРасскажи мне о себе: что тебе нужно для комфортного общения со мной? \nНапример, озвучивание текста, крупный шрифт или упрощенный язык.\nАня: Думаю, что мне понадобится распознавание речи.\nЭМПИК:\nОтлично! Я тебя понял. Включаю режим: <распознавание речи>Думаю, что мне понадобится распознавание речи.",
"setting": ["распознавание речи", "распознавание речи", {"технология": "text-to-speech", "принцип работы": "технология автоматического воспроизведения текста, например, функция “прямая речь” в iphone", "для кого": ["рас", "афазия", "нарушения процесса порождения речи"]}, ["рас", "афазия", "нарушения процесса порождения речи"]]
}
TODO
This class facilitates two scenarios:
Data Encryption:
Search for personal data indications in the knowledge base to encrypt them
Return the original text with encrypted data
Information Extraction for RAG:
Search the graph for relevant information and extract it
Use the extracted data to condition LLM