Ruchit Doshi Email & Phone Number

Irvine, California, US

• Researching on a novel AI solution for voice cloning in Mandarin and English language to generate Synthetic data using Text to speech models, reducing cost of outsourcing data collection by 20%
• Designing a state-of-the-art framework for modeling and training techniques to improve the computation efficiency by 18% for speaker verification project
• Collaborating with a cross-disciplinary team to build sustainable production models for Key word detection such as OK-GOOGLE, HELLO-BOAT, and ALEXA with state-of-the-art low FAR and FRR

• Worked as a core developer of a python-based Active learning framework aimed to reduce the labeling cost of customers by more than 20%
• Developed an MLOps platform capable of hosting customer’s BigData in parallel using Apache Airflow, Kubernetes, ElasticSearch and DynamoDB
• Handled customer use cases for object detection, NLP, image classification, and audio emotion detection using a flask-based SDK for model training and inference

Pittsburgh, PA, US

• Developed an open-source python package consisting of Data Analytics and Machine Learning models to help non-technical users use State-of-the-art techniques for automating the process
• Designed a ‘No-code’ end-to-end Machine learning pipeline development platform
• Achieved top 1% results for 10 datasets consisting of images, videos, and signals
• Presented the Automated Machine learning project at Neural Information Processing Systems conference 2020

Pittsburgh, PA, US

• Developed an Attention-based Deep Learning model using pyramidal Bi-LSTM network and CNN as a feature extractor for speech to text transcription
• Implemented various regularization techniques such as Teacher forcing, Gumbel Noise and weight dropout to increase the accuracy by 15%

Pittsburgh, PA, US

• Extracted vector of facial features by developing a model with skip connections to add model complexity and avoid vanishing gradient issue on a big data
• Performed Cosine Similarity and attained 93% AUC for the verification task
• Accomplished top 1% score on Kaggle competition with more than 300 colleagues

• Automated ETL processes across billions of data which reduced manual workload by 40%
• Optimized client’s logistics management by increasing the accuracy of the sales forecast by 10% using time-series predictive models- ARIMA and LSTMs

• Researched on a method for automatic reduction of fractured bone models derived from CT scans
• Implemented Iterative Point Cloud Algorithm to determine the position and orientation of the fragmented parts in 3-D space

Pittsburgh, PA, US

• Executed Feature Engineering on Paderborn University Bearing Dataset to classify raw vibration signals into healthy and unhealthy bearing classes
• Justified the results of PCA by comparing the accuracy of the entire dataset with the reduced feature dataset

Pittsburgh, PA, US

• Employed Scikit-Learn Algorithms such as Random Forest, Support Vector Machine, Logistic Regression, and XG-boost to achieve a recall of more than 98%
• Augmented the scope of the project by implementing Ensemble Technique to increase the accuracy of the shallow machine learning algorithms by 5 %

• Spearheaded a team of 20 members to design and fabricate subsystems of the KTM RC 390 engine. Followed a strategic approach and created an annual budget and timeline of the team to meet set goals

• Designed and Manufactured custom designed air intake system using the Selective Laser Sintering process
• Developed different engine tuning maps for various events in formula student competition by reviewing data of engine sensors using the data logger and telemetry module of Race-capture
• Led the presentation and guided the team to win the “Best Designed Formula Student Vehicle” at Formula Bharat 2017 and 2018 and won the ‘Cost AND Sustainability' event at Formula Student Austria, 2018

Master Of Science - Ms, Mechanical Engineering

Data Science Nanodegree, Data Science

Bachelor'S Degree, Mechanical Engineering