Research Paper

Core Innovation

This paper introduces Menta, the first small language model fine-tuned specifically for multi-task mental health prediction from social media using a LoRA-based framework and cross-dataset training. It achieves superior accuracy compared to larger models while enabling efficient on-device deployment with low memory requirements.

Why It Matters

Early detection of mental health conditions is critical but limited by the size and resource demands of existing models. Menta offers a lightweight, accurate solution that runs on mobile devices, enabling scalable and privacy-preserving mental health monitoring accessible to millions globally. This can transform mental health workflows by providing timely insights without compromising user data privacy.

Potential Customers & Pain Points

• Mental health clinics – Need scalable early detection tools
• Mobile app developers – Require lightweight privacy-preserving models
• Employers – Seek proactive employee mental health monitoring
• Insurance companies – Want improved risk assessment
• Researchers – Need accessible mental health prediction models.

Business Model

Licensing the Menta model to mobile app developers, healthcare providers, and insurers; offering SDKs and APIs for integration; potential subscription for continuous updates and support.

Competitive Landscape

• Ginger
• Woebot
• Wysa
• Mindstrong Health
• Spring Health

Implementation Challenges

• Regulatory approval for clinical use
• User trust and adoption of AI-based mental health tools
• Integration with existing healthcare systems
• Ensuring model accuracy across diverse populations

Validation Strategy

• Conduct clinical trials to validate prediction accuracy and impact
• Pilot deployments with mental health apps and clinics
• User studies to assess privacy perceptions and usability
• Benchmarking against existing mental health prediction tools