Varun Sharma

Enabled footage search for audits and operations with Python, YOLOv8, OpenCV and LLM workflows, reducing search time by 40%. Built Gemini/AWS Bedrock transcription and translation workflows with FFmpeg, subscription and admin features, supporting 500+ daily active users.

Prepared analysis and modeling datasets through Python/SQL collection, cleaning and validation pipelines, reducing data preparation time by 30% for stakeholders.

Enabled cited document retrieval with a LangChain/Pinecone assistant incorporating user feedback, improving retrieval accuracy by 15%. Added retrieval evaluation and latency monitoring, reducing system latency by 200ms; deployed the assistant on GCP with Docker and CI/CD.

Curated 10,000+ mammograms under HIPAA governance for a Drexel-Thomas Jefferson Hospital research collaboration. Benchmarked three architectures (ViT, Swin-T and CLIP) with patient-level splits and reproducibility controls, reducing model selection time by 2 weeks; reviewed limitations with clinicians.

Analyzed eight years of EHR/claims data covering approximately 250,000 patients and 120 features; compared models and assessed calibration, imbalance and SHAP explanations.

Delivered 15+ full-stack features across three regulated pharmaceutical systems using C#/.NET, SQL Server, JavaScript and Azure DevOps. Strengthened validation and exception handling, modernized publication approvals and automated build, test and deployment with QA/product teams, reducing deployment errors by 60%.

Grade: 3.91/4.0

Pursued a Master of Science in Data Science with a GPA of 3.91/4.0.

Earned a Bachelor of Technology in Computer Science and Engineering.