Mental Health Assistant
The Mental Health Assistant is a knowledge graph agent that reasons over a curated psychology and mental health dataset. It can answer clinical questions about disorders, symptoms, therapies, medications, neurotransmitters, and lifestyle risk/protective factors — and explain the connections between them.
Example questions it handles:
- "What are the best treatments for Post-Traumatic Stress Disorder?"
- "Which disorders are commonly comorbid with Major Depressive Disorder?"
- "What medications treat Bipolar Disorder I and are they FDA approved?"
- "What life factors influence Major Depressive Disorder?"
- "How does Sertraline work at a neurotransmitter level?"
Dataset and why a graph fits
The dataset was purpose-built for this hackathon covering 15 mental health disorders, 25 symptoms, 13 therapies, 14 medications, 5 neurotransmitters, and 10 lifestyle factors — with over 150 relationships between them.
A graph is the natural fit because mental health is fundamentally relational. A disorder doesn't exist in isolation — it shares symptoms with other disorders, responds to multiple therapies, is treated by medications that modulate specific neurotransmitters, and is influenced by lifestyle factors. These multi-hop relationships are exactly what graph databases excel at and what makes the agent's reasoning genuinely useful.
Relationships used:
(Disorder)-[:HAS_SYMPTOM {frequency, severity}]->(Symptom)
(Disorder)-[:TREATED_BY {efficacy_score, line}]->(Therapy)
(Disorder)-[:TREATED_WITH {line, fda_approved}]->(Medication)
(Disorder)-[:OFTEN_COMORBID_WITH {comorbidity_rate}]->(Disorder)
(Medication)-[:MODULATES {action}]->(Neurotransmitter)
(LifeFactor)-[:INFLUENCES {effect_size}]->(Disorder)
Tools used
- Cypher Templates — for structured queries (treatments, symptoms, comorbidities, medications, life factors, neurotransmitter mechanisms)
- Text2Cypher — for open-ended natural language questions
Note: full disorder names required (abbreviation support is a planned improvement)
Known limitations / future improvements
- Abbreviations like PTSD, MDD, GAD not yet supported — use full disorder names
- Similarity search (vector embeddings) not yet implemented
- Dataset could be expanded with more disorders and real-world prevalence data
Agent Tools:
Use of reasoning and thought:
Clean Response:
