Sentiment Systems
Built the NLP pipeline and multi-tenant backend that let clinics read patient sentiment without reading every entry.
What it did
Clinics collected patient feedback and had no practical way to read it. I built a sentiment analysis pipeline with spaCy, NLTK, and PyTorch, with custom preprocessing and topic clustering, so staff could see trends by topic instead of working through entries one at a time.
The backend
A Django admin dashboard with role-based access and Plotly-backed charts, which replaced manual review and cut reporting time by 30%. Underneath it, ETL pipelines in Pandas and SQL to clean genuinely inconsistent clinical records, and a PostgreSQL schema with Django ORM models designed so a new clinic could be onboarded without any chance of its data mixing with another's. REST endpoints via Django REST Framework, containerized with Docker, deployed on AWS.
Context
This was my earliest engineering role and the multi-tenancy design is the part I would still defend. Getting tenant isolation right at the schema level, before there were enough clinics for it to matter, meant onboarding never became the thing that broke.