Hi everyone!
I'm Mayki, a researcher at PGCOMP/UFBA and a member of RECSYS Labs. We are developing an advanced recommendation system that seeks to overcome rigid automation ("if this, then that"). Our focus is Multi-Label Incremental Learning: an AI that continuously learns from its use to predict complex states of multiple devices (brightness, temperature, speed) in an adaptive way.
What do you gain by collaborating?
By participating, you help validate AI models that prioritize user sovereignty and local processing. All results and code will be made available as Open Source, allowing the community to directly benefit from smarter state prediction algorithms. But this is just the beginning. This research is the basis for creating a proactive, local virtual assistant fully integrated with Home Assistant. While current assistants (Alexa/Google/Siri) are reactive, acting only on command, our goal is to develop an intelligence that learns from your routine to anticipate needs and interact organically. By collaborating, you help develop the "brain" of this future technology.
Why is your collaboration vital?
Most smart home datasets focus only on binary states (on/off). To create an AI that reflects the real complexity of a modern home, we need real-world usage data. The goal is for this technology to be Open Source and follow the principles of the Open Home Foundation: local execution, open standards, and user sovereignty.
How does the analysis impact the final result?
- Identification of Collective Patterns: The system learns to mediate the environment when people with different tastes occupy the same space.
- Prediction of Complex States: Trains the model to predict brightness levels, fan speeds, and climate control modes.
- Adaptation and Effort Reduction: The home becomes capable of recognizing changes in your routine and adjusting itself, becoming less dependent on repetitive commands.
Security, Privacy
We know that privacy is non-negotiable with Home Assistant:
- What we DO NOT collect: The dataset does not contain passwords, access keys, tokens, or network settings.
- Technical Filtering: The focus is on sensors and actuators. Although the HA database may record GPS events or voice transcriptions (TTS), our team performs technical processing to filter and ensure that only interactions relevant to the study are included, maintaining strict anonymity.
How to participate?
If you have a usage history of more than one month, your help will be fundamental to open science!
Export Guidelines:
For standard database users (SQLite)
For MariaDB users
Data Submission Form:
Link to the Participation Form
Academic Transparency:
Learn about my publications on Reinforcement Learning and Smart Homes: ORCID
I am available to answer any questions!
POST APPROVED BY @MissyQ