Saran Sundar

ADAPTIHIP

Design Endorsements

I came up with this in a hackathon held at VIT Vellore 2022. We were made to come up with possible design solutions in one night and I took this chance to modify the orthodox hip implant.

 

Our approach included an overall reduction in weight while maintaining structural stability, Inspired by the trellis bike frame

 

We removed the overall unnecessary materials in the hip implant by giving it a porous structural build. Even though it's theoretical and still under development.

 

Plant Disease Classification

Tiny ML - AI for Good by ITU

Trained the custom architecture for 7 epochs using augmentation techniques, achieving a remarkable 74.00% accuracy despite low parameters.

 

Leveraged the XIAO ESP32-S3's compact form factor, making it an optimal choice for discreet placement in diverse environments.

 

Acknowledging potential enhancements, the plan involves extending training sessions with an enterprise account to improve the model's precision and recall, secured 2nd place in the overall competition.

 

Wild Life Monitoring

Tiny ML - AI for Good by ITU

created a cost-effective TinyML solution for wildlife monitoring, focusing on reliable animal identification and biodiversity conservation.

 

Utilized XIAO ESP32-S3 and a MobileNet V2-based model, facing deployment challenges. Adopted the FOMO model from Edge Impulse for improved performance.

 

Achieved 66% precision and 50% F1 score with the FOMO model. Highlighted XIAO ESP32-S3's adaptability. Future plans involve extended training for enhanced precision and recall, showcasing a proactive approach to model improvement. Secured 3rd place in the overall competition

 

UBISOFT

Find The LAG

Participated in Find the LAG competition at Techfest, IIT Bombay by Ubisoft

 

The solution we came up with was to extract frame-by-frame images from videos and compute the structural similarity index measure on each frame against the frame when the input is given, we need not change the hyperparameters as feed in the input video, and it outperformed the traditional Background Subtraction Method

 

Secured 2nd place in the competition and received prize amount worth 40k

Chennai Cybersecurity Hackathon

Pixelated Footage Enhancement

Enhancing the pixelated CCTV footage, our approach was based on Res-Auto encoders.

High-quality videos were collected from the internet and several additional features were added such as gaussian noise, blur effect, erosion, and dilation to the extracted videos which were used as input and the original image as output

Our model attained 98.6 percent SSIM accuracy and helped us secure 3rd place.

Virtual commissioning

Endorsements

This process enables engineers to create a virtual replica of the system, integrating mechanical, electrical, and software components, and then test and optimize it in a risk-free, virtual environment. By doing so, potential issues and inefficiencies can be identified and resolved early in the design phase, leading to cost savings and ensuring that the real-world commissioning process runs smoothly

 

This letter of recommendation from my organization is a testament to my acclaimed skillset.

Star of Excellence Award

Endorsements

Was awarded as the star excellence student of the Mechatronics batch of the year 2023

 

Actively participated in several hackathons, took up several internships in different domains, and published research papers.

R Paper Pal

Endorsements

R paper Pal is an automized AI engine that eases researchers needs. It can recommend papers with similar context and translate them into users native Language

 

The model we used for Language Translation is MBART Large-50-one-to-many for multilingual machine translation. The Text Summarization part is done using cohere’s API