Hi, I'm
Data Engineer & Analytics Practitioner
Data Science graduate from the University of Maryland, College Park, with 3+ years of experience in data engineering. I enjoy building production-ready data systems, from real-time pipelines and scalable cloud infrastructure to systems that support applied machine learning and generative AI.
I’m a data engineer based in Dallas, TX, with a master’s in Data Science from the University of Maryland and a background in computer science, AI, and ML from SRM University. I enjoy building data systems that make it easier for teams to get the data they need and trust what they see.
My recent work has focused on scalable data pipelines and data quality. I’ve built pipelines with Python, PySpark, and AWS Glue, and used dbt and Snowflake to prepare data for reporting. At the UMD Counseling Center, I automated research data loads from six campus sources, cutting the wait for updated data from six hours to under 90 minutes.
I also enjoy working on applied machine learning projects. My research on deepfake detection earned a Best Paper Award at an international conference, and I placed in the top 10 out of 2,500 participants in a national computer vision hackathon while studying at SRM.
University of Maryland, College Park
SRM University
Efficient Deepfake Image Detection Using Dense CNN Architecture
Dallas, TX
A selection of data science and engineering work. View all on GitHub
Builds an end-to-end analytics workflow for travel products and operations, with dbt models in Snowflake, ThoughtSpot reporting, and read-only access through MCP.
Streams Coinbase market data through Redpanda into a Spark Bronze, Silver, and Gold lakehouse. PostgreSQL and dbt support data quality checks, reconciliation, and Power BI reporting.
Builds options volatility surfaces with SVI and forecasts volatility with GARCH(1,1). Uses VaR, CVaR, and HMM regime detection to analyze risk on a medallion data platform.
Benchmarked a distributed AI orchestration system (Claude + MCP) across Google Colab, AWS EC2, and AWS ECS Fargate under 50–100 concurrent users. ECS achieved 457 RPS with 50% fewer resources than EC2 (460 RPS), validating serverless containers for scalable agentic AI.
Analyzes Freddie Mac mortgage data at the loan level using validated pipelines for collateral analysis, loan performance, risk modeling, and constrained pool selection.
Designed a closed-loop neuro-symbolic pipeline combining LLM fluency with deterministic entity verification. Achieved 100% preservation of critical medical entities (dosages, vitals) across 300 texts vs ~99.7% for pure neural baselines (BART, T5).
Built a machine learning pipeline to predict patient/appointment no-show risk from EHR-style scheduling data using Random Forest and Logistic Regression models, enabling proactive scheduling interventions and reduced manual review effort.
Forecasts autonomous vehicle trajectories and classifies safety outcomes using the Argoverse 2 dataset.
Developed a binary deepfake image classifier using a Dense CNN architecture with data augmentation and dropout regularization. Leveraged TensorFlow and OpenCV for preprocessing and feature extraction, trained across 5 diverse datasets achieving 97% accuracy on unseen data. Published at IRCCTSD 2024.
A sensory companion app for autistic users, built with FastAPI and Claude AI.
Designed a scalable data lake pipeline on Databricks using Apache Spark and Delta Lake with Medallion architecture. Performed large-scale transformations via Spark SQL, enabling reliable, versioned, and query-optimized data for downstream analytics workloads.
Architected a production-grade ETL pipeline on Snowflake following the Medallion architecture (Bronze → Silver → Gold). Implemented data quality checks, schema evolution, and query optimization using SQL and Snowflake-native features for scalable analytics.
Analyzed EV adoption patterns using GeoPandas and Scikit-learn to map infrastructure gaps across Washington state. Built Random Forest and Logistic Regression models achieving 98.42% CAFV eligibility accuracy and R²=0.918 for electric range prediction.
Performed end-to-end retail analytics on real-world e-commerce data using MySQL covering complex multi-table joins, window functions, aggregations, and customer segmentation to surface actionable sales and product performance insights.
Explores trends in medical school matriculation using data from the Association of American Medical Colleges.
Explores diamond cut quality and its relationship to other diamond characteristics.
Uses sentiment analysis to explore audience reactions to TV shows.
Python
SQL
Apache Spark (PySpark, Spark SQL)
dbt
Airflow
Kafka
Redpanda
ETL/ELT Pipelines
Dimensional Modeling
Parquet
Pandas
SSIS
Snowflake
Databricks
Delta Lake
PostgreSQL
SQL Server
T-SQL
PL/SQL
Query Optimization
Amazon Web Services (AWS)
Amazon S3
AWS Glue
Amazon EMR
Amazon Athena
Amazon EventBridge
Azure ADLS Gen2
Azure Data Factory
Scikit-learn
PyTorch
Feature Engineering
Model Evaluation
Time Series Analysis
Hypothesis Testing
MCP
Amazon Bedrock
Azure OpenAI (GPT-4)
Claude
Cursor
FastAPI
MLflow
MLOps
Power BI
Tableau
ThoughtSpot
KPI Tracking
A/B Testing
Terraform
Docker
Git
GitHub Actions
CI/CD
Amazon CloudWatch
pytest
dbt Tests
Reconciliation
Data Lineage
Data Engineer · Remote, US
Data Engineer · College Park, MD
Data Engineer – AI/ML · Bengaluru, India
SQL Developer Intern · Hyderabad, India
International Research Conference on Computing Technologies for Sustainable Development (IRCCTSD 2024)
Communications in Computer and Information Science, vol 2361. Springer, Cham.
Deepfakes innovative manipulations of digital visual content using deep learning methods have emerged as a significant threat, raising concerns about misinformation and privacy violations. Their influence spans social media, political discourse, and beyond, highlighting the urgent need for robust detection tools. This research addresses the deepfake threat through a thorough examination of binary classification techniques. Focused on distinguishing genuine from manipulated images, the study leverages diverse datasets to train and evaluate methods utilizing Convolutional Neural Networks (CNNs) with emphasis on spatial feature extraction. Experimental results demonstrate the model's effectiveness in detecting manipulated images across various scenarios, achieving 97% accuracy on unseen data.
"Efficient Deepfake Image Detection Using Dense CNN Architecture"
Ranked in the top 10 out of 2,500 participants in Proglint's Alliance University Computer Vision Hackathon.
Download my full resume to see my complete experience, skills, and qualifications.
Download ResumeI'm always open to interesting conversations, collaborations, or new opportunities.
