Harun Gunasekaran

Built the Unified Student Record in dbt: one row per student per term, from their first term at Northeastern onward, joining census enrollment, admissions, degree outcomes and participation across seven experiential learning programs. On top of it sits a Snowflake semantic view and a Cortex Analyst agent that answers leadership's plain-English questions on enrollment, admissions, graduation, retention, degrees and program overlap. The guardrails are layered: the semantic view owns the SQL rules, the agent's instructions own scope and refusals, and joins that would quietly inflate rates are forbidden. It asks before guessing on an ambiguous year or cohort, never uses web search to back up an internal figure, and never identifies a student. dbt tests enforce the rules on every build, and every answer type is validated against an independent query. Validated and handed off, with the production rollout in progress.

Institution-wide analytics for Northeastern's enterprise Claude deployment. Built behavioral segmentation models and user archetypes; delivered adoption findings and engagement trends to senior leadership.

Architected and maintained Northeastern's production dbt transformation pipeline serving governed analytics marts across global education, honors, pre-matriculation, and co-op programs. Built executive Power BI dashboards translating enrollment, retention, and student lifecycle data into VP- and Dean-level insights.

Built end-to-end Tableau Prep pipelines to identify at-risk undergraduate students, delivering targeted intervention lists to student success teams.

Trained and deployed BERT-based transformer models to automate classification of co-op student experience reviews, replacing a fully manual process and enabling scalable program assessment.

Developed and deployed a collaborative filtering recommendation engine using AWS SageMaker. Paired with BERT-based NLP models for abstractive summarization, sentiment analysis, and multi-class text classification.