pdf2pixels
PythonA lightweight Python library that lazily renders PDF pages as PNG or JPEG images using interchangeable PDFium and Poppler backends.
Work
A selection of projects, research efforts, and deployed systems across industry and academia.
Conversational AI
Amazon
Self-Learning Conversational AI
Building conversational systems that adapt from interaction history, retrieval, personalization, and longer-term memory.
Alexa AIRetrievalPersonalizationMemoryEvaluation
Healthcare AI
Oracle
Pharmacovigilance AI
Working on healthcare AI for adverse-event understanding, evidence organization, and expert-facing safety review.
HealthcareLarge Language ModelsVision-Language ModelsInformation ExtractionSafety WorkflowsHuman-in-the-Loop
Privacy AI
Protopia
Privacy-Preserving Generative AI
Exploring generative AI approaches that preserve utility while protecting sensitive data.
PrivacyLarge Language ModelsGenerative AIEnterprise AITrust
Multimodal Systems
MetaOracle
Multimodal Content Understanding
Connecting language, vision, and extraction systems to understand content across platform-scale and multimodal healthcare workflows.
VisionMultimodal LearningContent UnderstandingInformation ExtractionMultimodal AI
Computer Vision
AmazonPhotokharmaVirginia Tech
Vision Systems in Real-World Workflows
Computer vision work spanning large-scale cloud inference, face detection and recognition, fraud detection, attribute prediction, and operational perception in messy real-world workflows.
Large-Scale InferenceFace DetectionFace RecognitionFraud DetectionAttribute PredictionReverse Logistics
Computer Vision
Virginia Tech
VIP: Finding Important People in Images
Understanding who matters in an image by reasoning about context, relationships, and scene structure rather than saliency alone.
VisionImportanceHuman-Centered AIScene Understanding
Open source
Tools
Focused utilities and developer tools I maintain.