Leading and developing AI Strategies for multiple products in the company. Implementing ML Product Management and Data Governance strategies to centrally govern multiple ML teams and large sources of dataWorking on Computer Vision, LLMs and predictive MLImplementing LLM and RAG pipelines in multiple projectsWorking with Diffusion models like Stable Diffusion and Flux for Image Generation Suite for Reliance Trends. Launched BharatDiffusion to create India's own image foundation models

Leading the research and engineering team of PixelBin, a A.I transformation and content delivery platform. Using the latest in LLMs and Image Generative Models to build cool products.→ Launched multiple AI solutions, including erase.bg, upscale.media, Diffusion Background Generator, Product Tagging, and image enhancement, achieving 250k daily traffic and 6.8M monthly visits and surpassing 1M$/year in ARR.→ Oversaw a research team of 7 in implementing CNNs, Diffusion models, and Transformers for various computer vision applications→ Managed a 4-person engineering team responsible for deployment infrastructure serving 250k daily users, optimizing scaling and cost with TensorRT and AWS Inferentia→ Established a 4-member 3D artist team leveraging Blender for synthetic data creation and 3D work-> Worked on a number of computer vision problems like super-resolution, object tracking, 3d vision and GAN based models-> Involved in deep learning research and optimising costs + lowering inference time of model deployments.

Fynd AML Research: research.fynd.comEraseBG: https://erase.bg - Core team member and principal researcher for EraseBG, a completely automated background removal service - Created data collection pipeline integrating active learning approaches- Lead the machine learning/deep learning/computer vision research required for the task of foreground identification and transparency prediction(Image Matting) - Built a robust solution from scratch that can handle a huge variety of cases - Took part in business decisions to create a Product Led Growth strategy for the product- Worked on object detection and tracking for retail store video analytics platform Fynd Trak - Implemented various metric learning approaches for visual recommendation search engine Read about what EraseBG does: https://bluesky314.github.io/bgremoval/2 Research Publications:- AlphaNet: An Attention Guided Deep Network for Automatic Image Matting IEEE publication: https://ieeexplore.ieee.org/document/9191371 Arvix Pre-print: https://arxiv.org/abs/2003.03613- Salient Image Matting (Currently in peer-review)Arvix Pre-print: https://arxiv.org/abs/2103.12337EraseBG is live on https://erase.bg Check out our ProductHunt launch: https://www.producthunt.com/posts/erase-bg

Intern at MeDAL(Medical Imagining, Deep Learning and AI Lab) →Worked on tumor segmentation from multimodal MRI data. Created and used extensions of dice loss to increase lower bound on model performance from 0.64 to 0.8 IoU in the presence of a distribution split to increase model reliability and trust in clinical settings→ Implemented representation learning models for automated feature extraction in tabular datasets for cases where data is anonymized or feature creation is not obvious→ Devised a loss function to fine-tune segmentation in histopathology images by decreasing noisy signals in the gradients of saturated models→Used distance maps to learn/incorporate multiple doctor annotations and messy samples→Applied and reviewed methods of Continual Learning to different medical datasets. See blog for literature review.

Automating image beautification and color corruption resolution

Understanding consumer trends using analytics and devising marketing campaigns. Worked on overall 8 projects including Aegon Insurance, Sin jeans,Big Bazaar and Wunderkids playschool