/ open problemsReturn to indexThe hardest problems are not model size. They are memory, consent, fidelity, inference, and trust.
Better user models and truly personalized, privacy-first AI will only come from creative research and engineering ideas.
We believe the future of AI is personal. These are some of the open problems we are working on at Onairos.
Sampling the highest-fidelity and least-noisy data about user preferences to reveal true underlying behavior instead of relying on noisy stated signals.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Anchoring user predictions and models to verifiable ground truth, whether from neuroscience studies like EEG and feedback, or continuous real-world usage patterns.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Drawing targeted inspiration from neuroscience where necessary to implement memory, personality, emotion, and decision systems that mimic and can model the human equivalent.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Detecting and filtering content misaligned with an individual's values or goals, so we can improve AI data quality and help users understand and adjust their digital habits. We have made strides in this space, adapting Meta's Tribe V2 into Neuroguard by Onairos, but the work is limited by the sample and training data behind Tribe V2.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Building composite, representative methods to carry a user's personality and digital self seamlessly across apps and experiences without losing fidelity. We pioneered this space with the Onairos data wallet, but more must be done to make it seamless, interoperable, and applicable to other domains like wearables. Apple Wallet is a good example of great UX and hints of interoperability.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Ingesting richer signals from wearables, biometrics, posts, comments, and passive behavior through new consent-driven loops to build more complete user representations.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Running powerful personal models on-device or in secure enclaves, delivering privacy-preserving, low-latency inference without compromising UX, battery, or capability. Traditional methods like differential privacy or homomorphic encryption may also be applicable.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Intelligently generating synthetic data to responsibly fill gaps in user history, temporal, behavioral, or contextual, for more robust and complete personal models.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
Verifying that input data and usage signals are free from bias, tampering, or manipulation through provenance, anomaly detection, and integrity mechanisms. Attribution and trust graphs are the most applicable solution so far.
If you have ideas here, are passionate about this, or have worked in this space before, join us and help change the world.
If you are building with Onairos, integrating the API, exploring collaboration, or thinking deeply about personal AI, reach out.