From Clinical Sessions
to Compassionate Responses
A transparent look at how we build an AI that actually understands mental health — not just simulates it.
The Clinical Data Pipeline
Anonymized Clinical Data
We partner with Lebanese clinics to collect therapy session transcripts. All data is rigorously anonymized — no names, no identifiers, no location data. Only the therapeutic dynamics remain.
The 'Ash' Difference: Fine-Tuning vs. Prompts
Unlike generic chatbots that just use a 'system prompt' to pretend they are therapists, My Hayat is fine-tuned directly on clinical datasets. Similar to advanced systems like 'Ash', it natively understands therapeutic questioning, emotional validation patterns, and crisis de-escalation because that's what it was built on—not just instructed to do.
Safety Audits & Validation
Every model version is tested against suicide/self-harm benchmarks, evaluated by our clinical advisory board, and compared against general-purpose AI for harmful response rates.
Your Conversation
When you talk to My Hayat, it uses an information-gathering stance — asking you thoughtful questions, not giving generic advice. It speaks your Lebanese dialect and understands your cultural context.
Pattern Recognition
Over time, My Hayat connects today's stress to patterns from weeks ago. Using NLP analysis, it discovers emotional patterns you can't see yourself and offers breakthrough insights.
Outcome Measurement
We track real-world outcomes: symptom reduction (PHQ-9, GAD-7), behavioral activation, and social support metrics. More engagement with My Hayat predicts better outcomes, not worse.
Real-World Results
Our proof-of-concept studies show measurable improvement across key mental health indicators.
What My Hayat currently does and what it doesn't claim
Conversations are not stored beyond the session by default
No account required for basic use
Safety classification runs before the language model
This is a prototype — privacy policy reflects actual implementation
