Manas Joshi Email & Phone Number

New Delhi, Dl, In

Toronto, Ontario, Canada

1. Strategic AI Project Management:A. Lead end-to-end AI project management, from conceptualization to execution, ensuring projects align with the company's goals and objectives.B. Collaborate with cross-functional teams to define project scope, deliverables, timelines, and resource requirements.2. Proof-of-Concept DevelopmentA. Develop and manage proof-of-concept projects to validate the viability and effectiveness of selected AI solutions.B. Collaborate with data scientists, engineers, and other stakeholders to design, implement, and test POCs.3. Risk Management and MitigationA. Identify potential risks associated with AI projects and develop proactive mitigation strategies.B. Regularly assess project progress, adjusting strategies and resources as needed to ensure successful outcomes.4. Knowledge Transfer and Training:A. Facilitate knowledge transfer sessions to educate teams on AI technologies, their benefits, and best practices for implementation.B. Provide training and support to ensure the adoption of AI solutions across various departments.

Japan

Project 1: Integration of Lightweight Shufflenet Backbone into Existing Head Pose Model1.The objective of this project was to integrate a lighter Shufflenet backbone into an existing head pose model. This would result in a low build cost and an efficient AI model that can run on low-powered hardware devices without needing a GPU.2. We used the Shufflenet backbone to reduce the model size while still maintaining a high level of accuracy. This made the model suitable for deployment in resource-constrained environments. This project has opened up new opportunities for developing low-cost, high-performance AI applications that can be used in a wide range of industries.Project 2: SSL Swin Vision Transformer Training using DINO, MoBY, and BYOL Methods1. The aim of this project was to train a strong, large vision model using SSL Swin Vision Transformer. We evaluated three different training methods, DINO, MoBY, and BYOL, to determine the best approach for customizing the model to our specific data.2. Using SSL training, we achieved superior results in terms of accuracy, robustness, and consistency. This has opened up new possibilities for the development of large-scale AI applications that can be tailored to specific use cases.3. This project has enabled us to provide our clients with more powerful and efficient vision transformers that can be trained on domain-specific tasks.Project 3: SSL Style Training of YOLOv5 CSP Backbone and Transfer Learning1. The objective of this project was to implement SSL-style training of YOLOv5 CSP backbone, a highly accurate detector that is robust and consistent in both production and development stages.2. Using SSL training, we trained the backbone via a vision transformer and then transferred it to YOLO to improve detection results. This project has enabled us to provide our clients with more options for customization. They can now train more powerful and generalized backbones on domain-specific tasks like person detection.

Toronto, Ontario, Canada

Master of Management in Analytics program, University of Toronto.

Toronto, Ontario, Canada

A major churn and sales loss is often based by retail stores during their renovation period. We built Power BI visualizations and ML predictive models to project sales based on various in-store and regional features. This allowed the team to proactively address potential challenges and capitalize on opportunities for growth during renovation period of a store.

Japan

Sapporo, Hokkaido, Japan

Gurgaon, Haryana, India