I came into analytics through business. That background shaped how I work: I start from the problem someone needs to solve, figure out whether the data can actually answer it, and make sure the person on the other end can use what I hand them.
MS in Applied Machine Intelligence at Northeastern University (4.0 GPA). Over the past year I've led first-author ML research with a six-person cross-institutional team, built evaluation pipelines for nonprofits across Maine, and designed databases reconciling messy federal data across 72,000+ census tracts.
I work across the full arc of a data problem: scoping, cleaning, modeling, validating, and communicating what it means to the people who need to act on it.
Designed a PostgreSQL database integrating four federal datasets (USDA, Census ACS, County Health Rankings, SNAP) across 72,531 U.S. census tracts. Built a composite vulnerability scoring model with statistical validation (p < 0.001). The SNAP gap analysis revealed that 45% of scored counties have safety nets that exist on paper but fail geographically. Interactive Tableau dashboards at state and county granularity.
End-to-end research pipeline from web scraping to a preprint on SportRxiv. No public dataset existed for NCAA penalty-kill analysis, so I built one from scratch: scraped 5 seasons of play-by-play data, integrated macro and micro datasets, and ran Q-learning and Markov Chain models to produce personnel deployment recommendations for coaching staff. Validated via LOSO cross-validation and 500-game bootstrap resampling.
Built a rule-based financial chatbot that analyzes Apple, Microsoft, and Tesla 10-K filings (FY2023 to FY2025). Extracted data directly from SEC EDGAR, engineered 12 derived metrics (margins, growth rates, cash conversion, leverage), and deployed a Flask web app with per-user session state. Caught and fixed a real normalization bug during testing where hyphenated queries silently returned wrong answers, and designed context-aware ratio explanations so the chatbot explains why a number is high or low, beyond simply stating it.
Completed as part of the BCG X GenAI Job Simulation (Forage), Aug 2026
Built segmentation models on 3,000+ survey responses for Wyman and Son (ME). PCA for dimensionality reduction and K-Means clustering with statistical validation (p < 0.001). Identified high-intent teen/tween households as the primary growth segment, directly informing client marketing strategy.
Designed and delivered an evaluation framework for healthcare and nonprofit programs across Maine. Combined Kirkpatrick survey methodology with CIPP analysis. Built a Python pipeline with TF-IDF text mining to audit qualitative stakeholder responses at scale. Report accepted by the LSHE team and briefed to program stakeholders.
If you could open only 10 new food pantries in America tomorrow, which 10 counties would save the most lives and why? The data to see all of this already exists. What didn't exist was a single reproducible ranking.
Read on Medium →
The hardest projects are the ones where there's no prior work to reference. No benchmark, no existing dataset, no paper to cite. Penalty-kill deployment in NCAA ice hockey was exactly that kind of project.
Read on Hashnode →Analyzed penalty-kill decision-making across 55 NCAA Division I programs using reinforcement learning and probabilistic modeling. Built the dataset from scratch (no public source existed), validated via leave-one-season-out cross-validation and 500-game bootstrap resampling. The framework gives coaching staff data-grounded personnel recommendations for situations that were previously guided entirely by instinct and film review.
Completed a simulation involving AI-powered financial chatbot development for BCG's GenAI Consulting team. Integrated and interpreted complex financial data from 10-K and 10-Q reports, employed rule-based logic and Python (pandas) to create a chatbot providing user-friendly financial insights and analysis. Deployed as a live Flask web app.
Completed a simulation focused on data analytics and commercial insights. Developed expertise in data preparation and customer analytics, utilizing transaction datasets to extract insights and deliver data-driven commercial recommendations. Identified benchmark stores for uplift testing on trial store layouts, enabling evidence-based decision-making. Created comprehensive reports for the Category Manager to support strategic decisions.
Worked 1:1 with a nonprofit partner to scope a pressing challenge, develop a solution, and present it at the Roux Entrepreneurship showcase. View badge →
I'm looking for full-time roles where data drives real decisions. If you're building something where the model needs to work and the person using it needs to understand why, I'd like to hear about it.
ravikumar.amr@northeastern.edu