Jan 2022 – Jan 2026 · 4 yrs · San Jose, CA

• Leading marketing quality content (text + images + videos) generation for Adobe GenStudio. Training VLM and diffusion models to evaluate, generate and fix visual content in an agentic fashion. Deploying them in an optimized fashion using vLLM.
• 13+ patents and 7 papers related to Firefly, GenStudio, Acrobat AI Assistant and other topics in Comp Science. More to come as we work towards building GenAI for enterprises ...
• CVPR paper: https://ui.adsabs.harvard.edu/abs/2023arXiv230305031R/abstract
• CVPRW paper: One shot target and shape aware video editing using diffusion
• Expert in ML and GenAI solutions like (controllable) image generation, prompt rewriting using LLM/VLMs, video generation, lipsync models using LatentSync, document cleaning using tesseract/ Restormer, on-brand models using custom LoRA/DoRA, custom scoring/eval models using large VLMs, business intelligence solutions like Snowflake and pipeline building solutions like Databricks, infrastructure solutions like RunAI, Modal, AWS/Azure, Baseten, strategy related to model moat, ML/LLM observability platforms like Arize, TrueEra, Datadog, industry trends in GenAI native SaaS related to upcoming startups like OpenAI, Anthropic, Midjourney, Runway, Hedra, BlackForestLabs, Morphic, HeyGen, ElevenLabs etc. I also pioneered the usage of personalized AI agents and do research on the no-code platforms which can help build superfast agents for your own use case. These include tools like n8n, Zapier and MindStudio. I also regularly delve into enterprise usage of no-code/low-code app builders like Lovable, Retool, v0 and Replit. I am also into analyzing the GenAI distribution platforms like Fal.ai/ Replicate, and comparing them with Firefly. I am also involved in continuous evaluation ai-dev-tools like Cursor, Windsurf, MS copilot and coupling Claude Code with Cursor/Windsurf etc.

Jan 2021 – Jan 2021 · 0 mo · San Jose, California, United States

• State-of-the art deep learning architectures for multimodal translation (text assisted image manipulation).

Jun 2020 – Aug 2020 · 2 mos · Manhattan, New York, United States

• Fast boundary integral solvers for PDEs. Wrote parallel high order solver for solving Laplace Beltrami PDE.

Jan 2014 – Jan 2014 · 0 mo · Bangalore

• Part of a team of 3 that developed a novel and a complete framework named EnTwine for
• identity resolution in online social networks and consequently creating a unified user prole
• by aggregating user information across networks
• Developed a new method for prior feature analysis using statistical methods instead of domain
• knowledge and heuristics
• Introduced a clustering step and provide two alternate ways to do it: one is a slight mod-
• ification of an existing clustering method called canopy clustering and other being a novel
• clustering algorithm in order to reduce space complexity of this step, both using results from
• the feature analysis step
• Hence providing a two phase solution to the identity resolution problem, first phase being
• candidate selection and the other being user identification
• Developed a new method based on Poisson Binomial Distribution and output probabilities of
• the classifier in the EnTwine framework to get an estimate of matching users

May 2013 – Jun 2013 · 1 mo · Mumbai Area, India

• Designed and implemented the mobile application for Purplle.com Salon and Spa Finder
• Part of team of 2 that designed and implemented the entire modular schema of the following features , using MVC architecture in Codeigniter Framework of PHP
• Completed Project Entailed Working on the following features –Venue Listing Refinement(Filtering), Reviews Module, Venue Tagging module and Automated Ad Tracking Module
• Optimization of MySQL Database using the best practices(Indexing, Slow Query optimization etc)
• Developed ERP module for Automated Ad tracking System and Venue Tagging