Kick-off Meeting



Add a caption...



Add a caption...

Resources:

• Standardised tests to measure psychotic symptoms are too cumbersome to administer routinely to everyone, can we infer them from clinical data using AI?
• Problem: How are you training/prompting LLMs to deal with psychiatric data?
• Need: It would be great to hear from people who are doing this!

• Help please with navigating upfront now - ethics and clinical trials and medical device regulation requirements as we are prepping our ethical approval process now. Be great to hear from other teams!

• What problem are you trying to solve?: Creating a culturally adaptive generative AI model for mental health screening (Depression, Anxiety, and Psychosis) in low resource settings of Kenya and Tanzania
• What could other teams help you with?: Understanding federated learning approaches. We would want to connect with teams working on culturally grounded models, to share notes.

• We have three main objectives: (1) develop an AI tool to assess therapy quality and generate culturally appropriate feedback for non-specialist counselors and supervisors delivering therapy in India; (2) evaluate its feasibility, acceptability, and preliminary effectiveness in improving therapy quality and patient outcomes; and (3) develop an ethical framework for the responsible implementation of AI-enabled supervision in routine mental health care.
• One area where we'd welcome input from other teams is evaluation. AI capabilities are evolving so rapidly that it can be challenging to design evaluation frameworks that remain relevant by the time a study is completed

• Problem we solve: RAISE addresses the lack of objective, scalable tools for identifying families at risk for intergenerational mental health problems before symptoms become chronic, and we hope to enabe learlier and more personalized prevention/intervention targets.
• Other teams: we have a special focus on using AI in realtime to learn more about emotional exchanges in real time so I think we can learn a lot from the AURORA team (as well as many others!)

• Develop culturally grounded generative AI tools to support para-counselors in Bangladesh who provide front-line mental health support for anxiety, depression, and psychosis.
• Responsible and ethical use of AI in low resource context.

• Building clinically faithful digital-twin patient agents by steering LLM activations with conceptors fit to real clinical interview data for training diagnosticians and robustness testing
• Need: shared standards for judging when a simulated patient is faithful enough to trust

• Problem: Depression fluctuates but care relies on infrequent, static self-report scales, so treatment adjustments come too late.
• Solution: A voice-first GenAI agent converses naturally with patients while collecting objective audiovisual and passive smartphone signals — combining GenAI engagement with deterministic ML measurement models for granular.
• Need: navigating investigational device exemptions in various country ahead of our RCT

• Problem solving: Scalable, Culturally-Informed AI for Fidelity Feedback and Adaptive Training in Youth Mental Health
• Need: Building AI-in-the-loop infrastructure while scaling.

• Problem solving: How can we augment and benchmark lay provider service delivery with AI?
• What could other teams help you with? Running evaluations and benchmarks.

• Problem solving: Preparing frontline crisis responders using a youth simulator and evaluator to enhance training and responder confidence
• Need: We would like to learn more about experiences with refining conversation simulation and approaches to providing quality feedback.

• Problem solving: We are developing ARIADNE which is a new voice-based interview system powered by generative artificial intelligence that can screen people for psychosis risk and generate tailored summary reports.  
• Need: We are eager to share experiences and learn from other teams regarding regulatory and translational challenges, as we believe we are likely navigating similar obstacles.

• Problem solving: Inadequate cultural validation of common used mental health tools in Nigeria.
• Need: We will look forward to help with creating an ethical and responsible AI framework as well as medical device regulation for real world implementation in Primary centers.



Add a caption...

Expanding work (with Gates) on evaluating frontier models for low-resource language understanding  https://gooey.ai/language-evaluation 
 Gooey.AI 's DPG-certification is imminent (meaning every dependency is open-source) Local install from  github.com/gooeyai  + paper at  https://gooey.ai/sovereignty 
Working on cross-cohort AI recipes
Plus, a bunch of new features (e.g. vibe code changes, easier UI)