Software Engineering Intern

• Built a PySpark reconciliation pipeline on Databricks pulling Workday API records into a bronze table and diffing them against the silver table, catching 302 employee and 2,860 position records missing from HR reporting.
• Released it to production through a change request with QA proofs and a rollback plan, scheduled as an Azure Data Factory job emailing exception reports to HR Information Systems twice monthly.
• Architected a company-wide data quality framework across bronze, silver, and gold layers with GPT-5.5-assisted rule generation, SCD Type 2 rule lifecycle, and error/warning severity tiers.
• Surfaced it all in a Databricks SQL dashboard the team still uses, and presented the framework to 50+ engineers and executives at final review.

Undergraduate Researcher

• Independently built BatteryGen, a 42M-parameter conditional VAE trained on 7.1M molecules in 16 hours on one RTX 3060, hitting 1.000 validity, 0.998 uniqueness, and 0.995 novelty against published baselines.
• Trained the paired prediction model on 575 zinc additives with molecular fingerprinting, RFE-CV column selection, Optuna tuning, and scaffold-grouped cross-validation, reaching R² 0.82 on Coulombic efficiency.
• Closed the loop with a multi-LLM reasoner, judge, and orchestrator validating final candidates against a retrieval-augmented literature base, advancing top additive NCNCc1ccccn1 to experimental validation.
• Shipped it as a pip-installable package with CLI entry points, gated by GitHub Actions running pytest on every push and auto-deploying artifacts to Hugging Face on merge.

Software Engineering Intern

• Built a document-processing pipeline extracting part specs from scanned oil-and-gas PDFs with Amazon Textract and Claude 3.5 Sonnet.
• Resolved near-miss matches against an Excel commodity-code dictionary using a lightweight local Llama model to reconcile abbreviation and wording variants.
• Loaded resolved parts to a stakeholder-facing CSV and PostgreSQL table through a Django application, cutting a 40-hour manual review to under one hour of human verification.
• Designed the output for human-in-the-loop review rather than full automation, surfacing every extracted field for correction before acceptance.

Engineer, via UTD Biotech Club

• Engineered an AI-powered wearable for real-time health monitoring, integrating physiological sensors with a microcontroller to collect and transmit biometric data.
• Developed Python ML models to analyze health metrics and flag cardiovascular events.
• Collaborated with a multidisciplinary team of engineers and clinicians on hardware and software integration.