Data Scientist
Statistical modeling, agentic AI, quantitative risk, and scalable data systems — turning complex data problems into measurable, real-world impact.
Intuitive
Data Scientist (Co-Op) • Sunnyvale, CA • Jun 2025 – Mar 2026
Framed a manual-support bottleneck as a data problem — combining EDA on usage logs with BERTopic clustering on ServiceNow & chat text to surface high-volume automatable workflows, then built agentic AI loops in LangChain with prompt-injection & data-leakage guardrails.
Trained a deflection classifier (logistic regression + gradient-boosted trees), measured impact through A/B tests & Power BI dashboards, ultimately automating 65% of support volume.
Discovery & Issue Mining
EDA on usage logs + BERTopic NLP clustering on ServiceNow & chat text to surface high-volume, automatable support workflows and quantify the automation opportunity.
Eval, Fine-Tuning & Guardrails
Built eval harnesses & benchmarks; fine-tuned HuggingFace/PyTorch models; diagnosed failed retrievals to improve grounding & answer quality. Agentic loops secured against prompt injection & data leakage.
Deflection, A/B Testing & Impact
Trained a logistic regression + gradient-boosted classifier to triage tickets. Ran rigorous A/B experiments with hypothesis testing (t-tests, proportion tests) to validate lift before full rollout — tracked via Power BI dashboards. Automated 65% of support volume.
University of Washington — IT
Data Analyst • Seattle, WA • Mar 2025 – Present
Analyzed usage & compliance data across university digital learning assets in Python & SQL, surfacing trends & KPIs that drove decisions on remediation effort and staffing.
Built an Adobe API integration into the data pipeline to auto-evaluate WCAG 2.1 compliance, generating accessibility metrics & reports so every student can access course content.
Educational KPI Analytics
Python & SQL analysis of digital-asset usage & compliance data across university systems, surfacing KPI trends that informed staffing and remediation prioritisation.
Automated Accessibility Compliance
Adobe API integration into the data pipeline auto-evaluates WCAG 2.1 compliance, generating accessibility metrics & reports for every digital learning asset.
Data-Driven IT Insights
Translated compliance and usage signals into actionable dashboards for IT leadership, reducing time-to-decision on remediation efforts and resource allocation.
ION Group
Quantitative Analyst • Pune, India • Jan 2024 – May 2024
Cleaned & analyzed historical market & position data (equities, rates, derivatives) feeding ION's XTP risk engine in Python (Pandas, NumPy), computing returns, volatility & exposures.
Independently recomputed VaR (historical & Monte Carlo), margin & P&L attribution as a golden-reference benchmark, then backtested exceptions with Kupiec-style tests and automated QA dashboards in Plotly.
Market Data EDA
Historical price & position data cleaning and analysis for equities, rates & derivatives; computed returns, volatility, and risk exposures feeding the XTP engine.
Risk Model Validation
Independent VaR recomputation (historical & Monte Carlo simulation) and P&L / margin attribution as a golden reference to benchmark and flag discrepancies in product risk analytics.
VaR Backtesting & QA
Backtested VaR against realized P&L using Kupiec-style exception counting (SciPy/statsmodels). Automated QA reports & Plotly dashboards replaced manual verification workflows.
Sequel2SQL
Microsoft Sponsored • University Capstone Project
Sequel2SQL is an agentic LLM + RAG framework engineered for automated SQL error diagnosis, rigorous optimization, and self-correction. Deep-dived into advanced SQL primitives — CTEs, Window Functions, query optimization — to develop nuanced reasoning methods for LLM & RAG performance.
By scrutinizing low-level execution-engine internals, we provided LMs with highly targeted semantic feedback, enabling a 6% absolute improvement over baseline on the BIRD-CRITIC Benchmark for NL-to-SQL tasks.
AST-Based Targeting
AST-based, segment-level parsing and transformation layers selectively process large queries — peak memory efficiency with highly targeted optimization strategies per query block.
Advanced SQL Execution
Low-level execution-engine mechanics translated into actionable reinforcement signals — query planner paths and database-native heuristics fed directly back to the LLM.
AWS — Scalable Data Systems
University Project
- Distributed data pipelines on AWS using S3, Glue, Athena, Redshift & Spark.
- SQL & Polars workflows with partitioning & index-aware schema design to optimize query performance across lake & warehouse layers.
- Benchmarked cost–latency trade-offs across storage types & compute models.
Distributed Data Processing Lab
Hadoop · Spark · Hive
- Configured an on-prem Hadoop/Spark cluster across 60+ Linux nodes with HDFS-based distributed storage.
- Spark ETL workflows for joins, aggregations & incremental data processing at cluster scale.
- Hive integration for structured querying over HDFS datasets with optimized partitioning strategies.
Early Roles
Where the fundamentals were built — production ML pipelines, cloud-scale ETL, and real-time edge inference.
UPL
Data Scientist Intern
Bangalore, India • Jun–Aug 2023
- Architected ETL pipelines on Azure Databricks, engineering PySpark feature pipelines on Delta Lake across 10+ years of time-series data for demand forecasting.
- Trained XGBoost, LightGBM & ARIMA models with time-series cross-validation; applied SHAP for feature selection & explainability.
- Predictive demand signal informed supply-chain decisions at scale across global agricultural markets.
Dronaid
Student — AI Head
Manipal, India • Jul 2022 – Jun 2023
- Curated large-scale drone image datasets in Python; ran EDA on class imbalance & active-learning sampling to select the most informative frames and lift performance on rare cases.
- Trained & evaluated CV detection models (precision, recall, mAP, error analysis); deployed the best on-device in C++ with CUDA for real-time edge inference.
- Achieved real-time performance on resource-constrained drone hardware through model compression & hardware-aware optimization.