Mental Health Facility Routing Pipeline
AAI 6600 · team project
A team prototype for finding care facilities; my contribution covered data collection, model evaluation, and clustering.
IllustrationHighlights
- Evaluated and tested open-source models for the needs-classification stage
- Built the data scraping and clustering that fed the pipeline (SAMHSA/NSUMHSS data)
- The wider pipeline scored 11,018 facilities (1,189 final), in a scikit-learn and sentence-transformers stack
How it was built
This course project routes people toward affordable mental-health care. Our team built a chatbot that classifies a person's needs, checks for crisis language, and matches them to facilities by location and insurance. My part was the data and evaluation work: I scraped the facility data from SAMHSA's treatment-facility survey, tested open-source models for the needs-classification stage, and built the clustering step that fed the pipeline. Teammates owned routing, scoring and anti-hallucination checks. The wider pipeline scored 11,018 facilities and kept 1,189. Limits: no per-person metrics are claimed.