Jun 2023 – Nov 2023 · 5 mos · Bengaluru, Karnataka, India · On-site

Sep 2022 – May 2023 · 8 mos · Gurugram, Haryana, India · On-site

Jun 2021 – Aug 2022 · 1 yr 2 mos · Bangalore Urban, Karnataka, India

• Hand Gesture Recognition:
• Utilized Keras Tuner to automatically identify the optimal set of hyperparameters for the CNN model.
• Fine-tuned pretrained models like Resnet50 & VGG16 on the gesture dataset to improve model accuracy.
• Achieved an average accuracy of 91% for 3 gestures and 82% for 5 gestures on the unseen dataset.
• Human Action Classification:
• Implemented custom data augmentations like Magnitude and Time Warp to improve the model generalization.
• Experimented with custom CNN and CNN-LSTM architectures and optimized them with hyperparameter tuning.
• Achieved an average accuracy of 87% and weighted average f1-score of 0.86 for 7 action classes

May 2020 – Jul 2020 · 2 mos · Bengaluru, Karnataka, India

• Deep Learning based CSI Compression:
• To create an efficient encoder and decoder modules to compress the CSI (channel state information) for
• transmission through the wireless channel and to easily retrieve it at the destination.
• Implemented an Autoencoder model with CNN block as encoder and Residual block as decoder
• Developed CsiNet-LSTM model reduced the NMSE from -17.3dB to -23.3 dB (6dB improvement) w.r.t CsiNet.