Published first-author ML researcher with a business foundation.
I build end-to-end pipelines that turn messy data into decisions people trust.
📊 Working on: a prior authorization / claim denial prevention tool using RAG over public CMS policy data.
🌱 Currently learning: Azure, agentic RAG architectures, evaluation frameworks for LLM-based tools.
💬 Ask me about: Q-learning, Markov Chain modeling, k-means clustering, or turning messy survey data into a segmentation model.
⚡ Fun fact: I came into data science through a business and management degree, so I care as much about the decision a model changes as the model itself.
Penalty-Kill Personnel Deployment and Offensive-Value Exposure in NCAA Ice Hockey First-author preprint · Q-learning (8-state reward structure, 10,000 iterations) + Markov Chain transition modeling, validated via LOSO cross-validation and 500-game bootstrap resampling across 55 NCAA D1 programmes |    |
Penalty-Kill Decision-Support Framework: Reinforcement learning system trained on 13,777 penalty outcomes across 55 D1 programmes. LOSO cross-validation, 500-game bootstrap, Wilcoxon p < 0.0001. Deployed as a live Streamlit research dashboard.
Yelp Business Closure Prediction: KNN classifier (75.2% accuracy, 89.2% specificity) on 11,266 businesses, paired with TF-IDF text mining and k-means clustering (k=7) across 10,000 reviews to surface qualitative failure patterns the structured model missed.
Maine Real Estate Price Analysis: Multi-predictor regression on 7,701 listings, improving explained price variance from 10.6% (single predictor) to 34.2% (adjusted R²) through feature engineering and property-type encoding.
Food Intervention Priority Index: PostgreSQL database integrating 4 federal datasets across 72,531 census tracts, with a composite vulnerability scoring model statistically validated at p < 0.001.
Consumer Segmentation & Motivation Analysis: K-means clustering and chi-square testing on 3,000 survey respondents, identifying a target segment with an 18.8 percentage-point higher purchase intent (p < 0.001).
Power Outage Analysis: Ten hypothesis tests on 65 weather-linked outage events, isolating wind speed as the strongest statistically significant driver of outage severity (p = 0.0044).
Languages & Databases
ML & Statistics
NLP TF-IDF · Cosine Similarity · VADER Sentiment · Text Mining
Visualization & Deployment
- Sports Research Analyst — AU,NCAA data partnership with Sacred Heart University, CT (Jan 2026 – Present): Built reinforcement learning and Markov Chain models on 13,777 NCAA penalty outcomes; published first-author preprint with a 6-author cross-institutional team.
- Data Analyst, Data for Social Good — Northeastern Roux Institute (Feb – Jun 2026): Designed a dual-phase AI evaluation framework for Maine nonprofits, combining Kirkpatrick survey methodology with NLP-based text mining.
- Ravikumar, A., Kaya, T., Artan, N.S., Taber, C., Morris, J.R., Raval, M.S. (2026). Penalty-kill personnel deployment and offensive-value exposure in NCAA ice hockey: a box-score decision-support framework. SportRxiv. doi.org/10.51224/SportRxiv.972
Finishing my Master's in Applied Machine Intelligence at Northeastern University, December 2026.