SOMESH ZANWAR

Arlington, TX · someshzanwar345@gmail.com · linkedin.com/in/someshzanwar · github.com/SomeshZanwar

Summary

M.S. Data Science candidate graduating December 2026 with professional experience in data analytics and hands-on work across AI/ML, analytics engineering, automation, API-backed applications, and agent governance. Experienced in Python, SQL, data validation, reporting automation, machine-learning evaluation, experimentation, and building production-minded systems with testing, controls, auditability, and technical documentation.

Education

Skills

• Programming & data: Python, SQL, pandas, NumPy, PostgreSQL, dbt, Excel, data transformation, validation, feature engineering
• AI & analytics: scikit-learn, XGBoost, classification, cross-validation, model evaluation, A/B testing, hypothesis testing, SHAP/LIME
• Engineering: FastAPI, SQLAlchemy, REST/JSON APIs, Docker, Git/GitHub, CI/CD, pytest, Streamlit
• BI & delivery: Power BI, Tableau, KPI analysis, requirements gathering, stakeholder reporting, technical documentation, Agile/Scrum

Experience

• Analyzed 80,000–150,000 operational records per cycle using SQL and Python; improved reporting accuracy by 30% through validation, reconciliation, and discrepancy resolution.
• Automated recurring reporting workflows, saving 4–6 hours per cycle; optimized SQL joins and incremental loads to reduce data-preparation time by 40%.
• Collaborated with three internal teams on Power BI dashboards, KPI definitions, reusable datasets, and recurring reports.
• Documented metric definitions, transformation logic, validation procedures, and discrepancy resolutions to improve consistency and maintainability of recurring reporting.

Projects

• Developing a data-driven workflow with UTA's Office of Information Technology to analyze incoming support tickets and improve routing to appropriate service teams.
• Applying Python-based data preparation, text/feature analysis, model evaluation, and reproducible engineering practices within a real operational support workflow.

• Built an evidence-bound execution control layer for autonomous analytics agents, enforcing policy at the action boundary; used signed permits and explicit authorization evidence for consequential actions.
• Implemented 21 automated tests covering policy logic, permit validation, and HTTP behavior; designed reviewable failure paths so unauthorized or insufficiently supported actions are blocked before execution.

• Built an operational data-trust system with 12 automated quality checks, incident lifecycle tracking, downstream lineage, automated CI validation, and trust reporting.
• Connected data-quality failures to downstream impact and supporting evidence, turning failed checks into reviewable incidents rather than allowing invalid data to propagate silently.

• Built an experimentation workflow over 50,000 simulated users and 280,000+ events, calculating conversion lift, p-values, confidence intervals, and segmented results.
• Translated statistical evidence into rollout recommendations, connecting experiment results and segment behavior to practical product decisions.

Open source

• Microsoft Agent Governance Toolkit: 5 merged PRs spanning policy/data-quality integration, a dbt evidence adapter, AgentMesh scaffolding, and tutorial/documentation fixes.
• AgentTrust / A2A CTEF: contributions across verifier enforcement, RAG scan states, runtime support, identity documentation, and agentic-security evidence; proposed a data-quality trust scheme with three conformance vectors.