Member of Technical Staff 1 at Adobe

VerifiAI (AI Content Authenticity & Detection): Engineered a hybrid detection pipeline for AI-generated and modified media by integrating forensic signals (noise patterns, compression artifacts) with ML classifiers based on Diffusion Reconstruction Error. Achieved 95% accuracy and aligned outputs with Adobe’s CAI/C2PA standards to establish a verifiable trust layer. Telemetry Logging & Insights: Designed Illustrator’s in-memory telemetry logging stage to capture 15,000+ real-time performance events daily without degrading speed. Integrated with DatabricksforSQL/Python analysis, uncovering insights (e.g., only 32% of edits ran at 60 FPS). These findings drove Roadmap changes that improved smooth operations to 70%, aligning with customer feedback. Integrated Infinite Resolution Textures: Adapted Nvidia’s approach on infinite resolution textures to Illustrator zoom operations, enabling super-resolution rendering across 80% of workflows and improving performance on large canvases.

Member of Technical Staff 2 at Adobe

Architected a plug-and-play agentic system translating natural language prompts into fully editable Illustrator effects, with a context-aware conversational interface, state tracking, and client-side validation to ensure reliable outputs from LLMs (GPT-4, Gemini, Claude) and VLMs (BLIP-2, Kosmos-2). This cut effect search and application time by 50%. Spearheaded development of agentic RAG system indexing internal codebases into a vector database, cutting codebase search time by 30% and enabling more effective context-aware queries across the org. Enhanced Illustrator transform operations with advanced graphic caching and tessellation, achieving a 5-12×frame-rate boost and improving performance for 87% of users.