Hi there!
Welcome to my portfolio.

Built a full‑stack data cleaning web application that automatically detects and fixes common data quality issues in CSV and Excel datasets. The platform performs duplicate detection, date normalization, numeric standardization, and configurable missing‑value handling while providing before/after previews of the cleaned dataset.

• Developed a FastAPI backend using Pandas to perform automated dataset profiling and cleaning.
• Implemented configurable cleaning modes including strict removal, safe retention, and automatic value filling.
• Built a responsive frontend interface supporting drag‑and‑drop uploads and interactive preview tables.
• Deployed the backend on Render and the frontend on Vercel to provide a publicly accessible data cleaning tool.

Architected and deployed a serverless, end-to-end ELT framework on Databricks to quantify the financial impact of social media marketing. The system integrates real-time platform metrics with enterprise-scale clickstream data to enable granular ROI analysis for Influencer and Native content.

• High-volume ingestion: built automated connectors for the YouTube Reporting & Analytics APIs and merged with large-scale internal clickstream logs.
• Delta Lake architecture: built robust PySpark pipelines with ACID transactions and idempotent merge logic for schema evolution and backfills.
• Data quality: enforced schema checks and null constraints prior to commit.
• Security + optimization: integrated OAuth 2.0 via Azure Key Vault and used serverless compute to optimize cost.

Built an end-to-end machine learning system that automatically measures surface tension from droplet images in under 1 second — 1000× faster than conventional methods.

• Developed a physics-based synthetic data generator using Young–Laplace equations.
• Trained a 5-layer deep neural network on 10,000+ droplet shapes.
• Implemented a complete computer vision pipeline for real-time image processing.
• Replicated published research results from Kratz & Kierfeld.

Designed and deployed a production-grade fraud detection system achieving ~4ms average inference latency while maintaining strong fraud-detection performance across 200+ transactions per second.

• Architected a streaming ML pipeline using Kafka for ingestion and Redis for low-latency feature caching.
• Implemented real-time features capturing transaction velocity, geolocation anomalies, and temporal patterns.
• Handled imbalance using SMOTE and cost-sensitive learning.
• Built monitoring with a dashboard and containerized deployment using Docker.

Built an AI-powered analytics assistant that converts natural language questions into executable SQL queries and returns structured insights through an interactive dashboard.

• Designed a full-stack architecture integrating Streamlit UI, FastAPI backend APIs, OpenAI LLMs, and DuckDB.
• Implemented a natural-language-to-SQL pipeline that transforms business questions into executable database queries.
• Built interactive dashboard components that visualize campaign metrics, revenue comparisons, and query results in real time.
• Created an end-to-end AI analytics workflow bridging natural language questions, structured data queries, and automated insights.

Designed and built an end-to-end machine learning platform to predict customer churn and forecast 90-day revenue for a simulated usage-based SaaS product.

• Generated 400K+ events across 2,000 accounts and ingested them into a Dockerized PostgreSQL warehouse.
• Built behavioral features across usage intensity, latency, errors, payment failures, and support activity.
• Created forward-looking labels designed to avoid data leakage.
• Trained XGBoost models under heavy class imbalance.

Worked as a Data Engineer on DataMAx, a cloud data warehouse and integration product supporting ingestion, profiling, cataloging, quality, provisioning, and governance for large-scale DWBI needs.

• Designed warehouse models across pharmaceutical domains.
• Built and optimized pipelines using Azure Data Factory, Databricks, and Python.
• Worked across Snowflake and Redshift for schema design and performance tuning.
• Developed validation and monitoring frameworks plus production-grade data services via FastAPI.

Engineered the end-to-end data path for an autonomous perception stack, including synchronized camera and LiDAR ingestion, dataset generation, and model packaging for reliable real-time inference.

• Built ingestion services for multimodal sensor data and aligned timestamps for training and inference.
• Created reproducible dataset pipelines with metadata, versioning, and standard splits.
• Packaged models for low-latency inference with health checks and fallback logic.
• Established validation gates and drift monitoring.