Projects

Problem: Predict no-shows + right-size scheduling for a virtual clinic.

Showcases: Feature engineering (calendar, weather, message sentiment), XGBoost, cost-sensitive metrics, optimization for overbooking, explainability (SHAP), fairness checks.

Deliverables: Notebook, REST API (FastAPI), small dashboard (Streamlit), and ethics card.

Problem: Reduce reactive overtime/temp-staffing costs by forecasting demand and right-sizing staff schedules in advance.

Showcases: Multi-source data integration (staffing, scheduling, operational systems), weekly time-series dataset construction (SQL), scenario-based demand forecasting (expected/high/low volume), cost modeling (overtime vs. temp labor vs. backlog cost).

Deliverables: Consolidated weekly dataset pipeline, forecasting model (Python/R), staffing-cost trade-off dashboard.

Problem: Operational databases lacked common identifiers, causing inconsistent records and low trust in reporting.

Showcases: ETL pipeline design (SQL/SQLite), probabilistic/fuzzy record matching across disparate sources, anomaly detection, data provenance, and limitations documentation.

Deliverables: ETL pipeline, record-reconciliation mapping logic, data quality scorecard, stakeholder-facing data dictionary.

Problem: Leadership needed to understand why users disengaged during onboarding to prioritize the product roadmap.

Showcases: Mixed-methods research design (Qualtrics surveys + usability interviews), A/B testing, qualitative coding and synthesis, cross-functional collaboration (Asana/JIRA).

Deliverables: Research report, synthesized insight summary, executive-ready recommendation deck.

Problem: Leadership lacked a single, reliable view of performance across siloed operational systems.

Showcases: Dimensional modeling, cross-system KPI alignment and validation, dashboard design (Power BI/Tableau), executive-ready data storytelling.

Deliverables: Power BI dashboard, data model documentation, KPI glossary.

Problem: Build a solver to generate semester plans that meet degree + visa constraints.

Showcases: Constraint programming (OR-Tools), explainable schedules, UX for “why this plan.”

Deliverables: Web app where students toggle constraints; export to calendar.

Problem: Evaluate selection models for bias; propose debiasing.

Showcases: Metrics (EO, DP, EOdds), propensity score reweighting, counterfactual evaluation, governance docs.

Deliverables: Repro audit report + sliders to see fairness/accuracy trade-offs

Problem: Recommend kid-friendly weekend events, factoring in cost, weather, and drive time

Showcases: Retrieval + ranking, graph of venues/users, weather features, travel-time constraints; LLM summaries.

Deliverables: Live mini-app + “Autobook” button stub to calendar.

Problem: Predict claim cost + flag suspicious patterns.

Showcases: Tabular ML + calibration, partial dependence, cost curves; human-in-the-loop review.

Deliverables: Analyst playbook + adjudication UI mock.

Problem: Predict drop/underperformance; recommend interventions.

Showcases: Time-varying features, uplift modeling, causal notes (DID/PSM).

Deliverables: Intervention planner + impact dashboard.

Problem: Build a testbed to score prompts/RAG pipelines on precision, hallucination, latency, cost.

Showcases: Eval datasets, automatic judges + human labels, confidence scoring, guardrails.

Deliverables: Reusable harness + leaderboard + doc.