Kumar Karthik
Ankasandra Naveen

CareEscapes AI is a dental tourism platform serving 10,000+ patients across 50+ countries. The platform needed a personalized recommendation engine to match patients with the right providers, and a data-driven framework to understand where patients were dropping off in the booking funnel.

Built a collaborative filtering recommendation engine using Matrix Factorization (SVD) via the Surprise library and Scikit-learn, generating personalized product analytics insights across 50+ provider networks. Designed an A/B test framework in Python and SQL to evaluate visit-to-booking conversion gaps, and delivered a Tableau dashboard tracking model performance and booking trends.

The A/B framework identified statistically significant conversion gaps at the provider selection stage, directly informing pricing strategy and platform UX recommendations for the product team.

J Healthcare Initiative, an NGO focused on reducing overdose deaths across the US and Canada, needed a data-driven tool to guide clinic placement decisions. Overdose data existed across fragmented CDC sources, making it difficult to identify geographic and demographic risk patterns at scale.

Consolidated 100,000+ opioid overdose incidents (2023–2025) by building a Python and SQL pipeline to ingest CDC records via API, producing a clean dataset across 7 opioid types and 5 demographic groups. Designed a multi-view Tableau dashboard with a US geospatial heat map, demographic breakdowns, and risk factor comparisons.

The dashboard surfaced distinct regional clustering of overdose risk by opioid subtype and demographic group, enabling NGO leadership to make evidence-based clinic placement decisions across the US and Canada.

Bright Mind Enrichment operated across 8+ U.S. locations but relied on static spreadsheet reports for program performance tracking. Data preparation was manual, error-prone, and consuming 10+ hours of staff time weekly with no real-time visibility for leadership.

Engineered automated ETL pipelines using RESTful APIs and Pandas to extract and consolidate program engagement metrics across all locations into a centralized reporting layer. Built 3 interactive Tableau dashboards tracking real-time KPIs for senior leadership, and automated daily report generation and distribution using Python scheduling scripts.

Reduced data preparation time by 60% and eliminated 10+ hours of weekly manual reporting, enabling program managers to shift from data wrangling to decision-making across all locations.

Manual interpretation of handwritten prescriptions is a major source of pharmacy errors. The goal was to build an agentic AI system that could automate prescription parsing and integrate with real-time inventory management — without hallucinating drug names or dosages.

Engineered a LangGraph node-based reasoning flow (Parsing → Validation → Decision) constraining a GPT agent to parameterized SQL operations, eliminating hallucinations and unauthorized inventory queries. Deployed the system as a live Gradio application on Hugging Face Spaces, enabling real-time prescription interpretation with structured drug name, dosage, and frequency outputs.

The constrained LangGraph architecture successfully prevented hallucinated drug entries — a critical safety requirement — while maintaining accurate real-time interpretation of handwritten prescription inputs.