Rahul Kadam Email & Phone Number

India

Improving ASR performance. Integrating Large Language Models into production.

Scaling Automatic Speech recognition to process 3000 audio files in a day. Implemented realtime ASR with 2 sec latency. Improved ASR accuracy via post processing

1. Automatic Speech Recognition: Implementing NeMo-based multilingual speech processing pipelines.2. Voice Activity Detection: Developed a model to reduce the latency of voice activity detection from 5 minutes to 2 Secs.3. Ship Detection: Implementing a ship detection model for satellite images using Yolov7 and Yolov5.

Delhi, India

1. Research Paper: Implemented Re-Identification with Consistent Attentive Siamese Networks Paper for an internal use case.2. End to End Use case & Tools: Deployed Black polythene detection use case to the production using Docker & Linux.3. R&D Team: Worked closely with RD Team towards improving model accuracy and precision for CCTV cameras

India

1. Dimensionality Reduction: Worked on Dimensionality Reduction of Human Activity data recorded using sensors and reduced the 561-D to 2-D to get the visualization of data and also calculated Eigenvectors to learn more about data.2. Sentiment Analysis: Worked on Sentiment Analysis using the raw dataset from Twitter and categorized them into 5 classes.

1. Detection Model: Build a Vehicle Number Plate Detection model with more than 95% accuracy.2. Neural Network: Trained a YOLOv4 architecture with PyTorch and thoroughly explored the architecture of the neural network.3, YOLO: Worked on training various NN models for custom object detection with YOLOv3 for different use cases like Seat Belt detection.