Inspiration
Mental health support often begins only after someone actively asks for help. But people experiencing prolonged distress may not always recognize, communicate, or report what they are going through.
We wanted to build a system that could identify changes in emotional wellbeing earlier, continuously monitor distress, and help connect people with appropriate support.
This led to AMMA — AI-powered Mental Health Monitoring Assistant, designed around one simple idea: support should be proactive, not just reactive.
What it does
AMMA provides continuous, multimodal mental-health monitoring using text, voice, behavioral, and engagement signals.
It uses NLP, sentiment and emotion analysis, and voice-based features to identify changes in a person's emotional state. These signals are combined by a Dynamic Distress Score (DDS) to represent the individual's current level of distress and track how it changes over time.
When concerning patterns are detected, AMMA can support an early-intervention workflow, helping route cases toward appropriate human assistance such as counselling, medical support, legal assistance, or protection services.
Rather than replacing mental-health professionals, AMMA acts as an early-warning and monitoring assistant.
How we built it
AMMA follows a modular architecture:
User → Voice/Text Input → AI Analysis → Distress Fusion Engine → Counsellor Dashboard → Intervention
- NLP & Sentiment Analysis — analyzes textual communication and emotional indicators.
- Emotion Engine — identifies potential emotional patterns from multimodal inputs.
- Voice Analysis — extracts relevant non-content speech characteristics.
- DDS Fusion Engine — combines multiple signals into a dynamic 0–10 distress score.
- Longitudinal Monitoring — tracks distress trends instead of relying on a single interaction.
- Counsellor Dashboard — provides trends and alerts for human review.
- Intervention Layer — facilitates escalation toward appropriate support channels.
The system is designed with a human-in-the-loop approach, where AI assists professionals rather than making sensitive decisions autonomously.
Challenges we ran into
One of our biggest challenges was determining how to convert complex and subjective emotional signals into a meaningful, interpretable distress indicator.
We also had to consider:
- Combining different modalities such as voice and text.
- Reducing false positives and false negatives.
- Maintaining user privacy while handling sensitive information.
- Making AI-generated assessments understandable to human professionals.
- Designing an architecture that could eventually operate reliably at scale.
- Ensuring that the system supports professionals instead of attempting to replace them.
Accomplishments that we're proud of
We are proud of creating a multimodal mental-health monitoring architecture that goes beyond a conventional chatbot.
Our key achievements include:
- A Dynamic Distress Score for continuous monitoring.
- Multimodal analysis combining text and voice signals.
- Longitudinal distress tracking.
- A counsellor-oriented monitoring dashboard concept.
- An early-warning and escalation workflow.
- A modular architecture that can integrate different AI models as the system evolves.
- A human-in-the-loop design for sensitive mental-health decisions.
Most importantly, we transformed the idea from simply detecting emotions into a system focused on detecting changes in distress and enabling earlier human intervention.
What we learned
We learned that building an AI system for mental health is not simply an ML problem.
It requires equal attention to privacy, explainability, bias, human oversight, reliability, and responsible AI.
We also learned that a single AI prediction is rarely enough for sensitive applications. Tracking patterns and changes over time can provide more useful context than looking at one isolated interaction.
Most importantly, AMMA taught us that AI should augment human care rather than replace it.
What's next for AMMA
Our next step is to move from prototype to a more robust and clinically responsible system.
We plan to work on:
- Improving multimodal AI models.
- Personalized and longitudinal distress baselines.
- Better false-positive/false-negative handling.
- Real-time monitoring and alerts.
- Secure data storage and privacy-preserving AI.
- Explainable distress indicators for counsellors.
- Integration with voice, SMS, chatbot, and mobile interfaces.
- Extensive testing with domain experts and ethically approved datasets.
- Stronger human-in-the-loop intervention workflows.
Our long-term vision is for AMMA to become a proactive mental-health monitoring assistant that helps identify distress early and connects people to the right human support at the right time.
Built With
- antigravity
