Naveen Raju S G

Graduate Teaching Assistant

Illinois Institute of Technology, USA

• Job responsibilities include supporting faculty in classroom instruction, facilitating discussion sessions, holding office hours for student consultations, evaluating assignments and exams, supervising exams, and offering constructive feedback on assignments.

• As a Graduate Teaching Assistant for the Data Mining course, I handled a class of 93 students.

Engineer CL2-I

Samsung Electro-Mechanics Software India Bangalore Private Limited, India

• Worked on a deep learning based number plate detection and character recognition algorithm with overall accuracy of 90%.

• Improved the accuracy of crowd detection and human trespass detection algorithms by 25%.

AI Engineer

Telerad Tech Pvt Ltd., India

• Led a team of 4+ using the Agile software development method, including project planning, road map creation, release planning, sprint planning, and constant client interaction; assisted in developing proof-of-concept prototypes; removed project development bottlenecks; and instructed team members on technical topics such as deep learning and computer vision.

• Customized a U-Net based architecture to segment different types of lymph nodes in mammograms with specificity and sensitivity of 90% each.

• Developed a deep learning-based architecture and image processing logic to detect asymmetry in pairs of mammograms with an F1 score of 95%.

• Built an image processing algorithm and CNN classifier to detect and reduce false positives of micro, macro, and amorphous calcification in mammograms with recall of 92% and precision of 90%.

• Developed custom CNN architectures for segmentation and detection of various lung conditions, including Cardio-thoracic ratio, Pleural effusion, consolidations, and Pneumothorax, utilizing image processing techniques for localization and quantification of the detected regions with a dice score of 95%, a MAP score of 90%, and an overall accuracy of 93%.

AI Intern

Telerad Tech Pvt Ltd., India

• Customized deep learning architecture for mammogram lesion segmentation, combined with image processing logic to improve specificity and sensitivity. Obtained an overall IOU score of 94% and an F1 score of 92.5%.

Fine-Tuning the FLAN T5 LLM Model for Enhanced Dialogue Summarization

• A Comprehensive Approach with Full Fine- Tuning and PEFT, Evaluated Using ROGUE Metrics.

Enhancing Resume Shortlisting and QA using LangChain, LLM and Retrieval Augmented Generation RAG

• This project endeavors to transform the conventional resume shortlisting process by tackling prevalent issues like keyword-centric focus, limited context understanding, and uninformed skill assessment within traditional ATS software. Through the integration of cutting-edge technologies such as Large Language Models (LLM), Retrieval-Augmented Generation (RAG), and LangChain, the initiative showcases several potential solutions to elevate contextual analysis, skill assessment, and portfolio utilization. The primary goal is to create a more accurate and equitable representation of candidates' qualifications, thereby enhancing efficiency and fairness in the hiring process.

Enhance Positive Summary Generation by Fine-Tuning FLAN-T5 through Reinforcement Learning

• Used a reward model that predicts either "not hate" or "hate" for the given text and also used Proximal Policy Optimization (PPO) to fine-tune and detoxify the model.

Statistical Analysis and Modelling of Real-Estate Price Prediction

• This project explores the concepts of statistical data analysis and modeling various regression techniques to predict the cost of occupied homes. The objective is to analyze housing data and develop accurate regression models for home price prediction.

Finding the pattern behind the online shoppers purchasing intention

• Objective is to analyze trends in the online shoppers purchasing intention dataset using exploratory data analysis techniques, and build machine learning models to predict whether a new customer is likely to make a purchase based on their browsing and purchasing behavior.

Real-time Machine Learning inference system using AWS cloud services to predict Taxi ride fare

• Trained and deployed a real-time ML inference system to predict New York taxi ride fares using Amazon web services: Amazon SageMaker, Amazon Kinesis Data stream, Amazon Kinesis Data Analytics, Amazon API Gateway, AWS Lambda, Apache Flink, Cloud 9 and S3 bucket.

Automated end-to-end ML workflow for fraud detection in auto insurance using AWS cloud services

• Processed raw data using SageMaker Processing jobs to create training, validation, and test splits in S3.
• Trained an XGBoost model with SageMaker training jobs, storing the trained model artifact in S3.
• Evaluated model performance on the test dataset using SageMaker Processing jobs, saving the evaluation report in S3.
• Conducted conditional checks on model performance and initiated predefined steps in SageMaker Pipelines accordingly.
• Utilized SageMaker Pipelines to create and register the model, check for bias, and generate model explain ability reports in S3.
• Deployed the model to a SageMaker Real-Time Inference endpoint using predefined steps in SageMaker Pipelines, enabling fraud prediction for auto insurance claims.

Creating a Search Engine with TF-IDF Algorithm for Wikipedia Data using Apache Spark on AWS EMR Studio to Analyze and Rank Relevant Documents

• Leveraged AWS EMR Studio and PySpark to develop a Search Engine, applying TF-IDF algorithm for analysis and ranking of Wikipedia documents.

Implementing a Streamlined Real-time Data Streaming and Analytics Pipeline with AWS Services

• Designed and executed a streamlined real-time data streaming and analytics pipeline with AWS Kinesis, Firehose, Data Streams, Kinesis Analytics Application, Glue Crawler, Glue ETL, and Athena for querying S3-backed databases.

Forex data pipeline using Apache Airflow

Developed a Directed Acyclic Graph (DAG) for Forex data pipeline using Airflow.
Directed Acyclic Graph (DAG) includes:
  • Check availability of forex rates
  • Check availability of file having currencies to watch
  • Download forex rate with Python
  • Create a Hive table to store forex rates from the HDFS
  • Process forex rates with Spark
  • Send a Slack notification
  • Add dependencies between tasks

Image classification using hierarchical based shifted window Vision Transformers

• To demostrate how SWIN Transformer performs efficiently compared to Vision Transformer(ViT) due to hierarchial processing with shifted windows and improved long range dependency handelling. Both models where trained on subset of Food-101 dataset.

Identification of Salt Regions in Seismic Images using Deep Learning

• Using Seismic Images to segment salt deposits beneath the Earth's surface using Deep learning based segmentation algorithms.

Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks

• Zebra to Horse and vice versa image to image translation was achieved by training Cycle-Consistent Adversarial GAN with custom designed generator and descriminator.

• Pneumothorax Detection and Classification on Chest Radiographs using Artificial Intelligence - Lattice the Machine Learning Journal Volume-2 Issue-1, ISSN 2582-8312
• Pneumonia Detection and Classification on Chest Radiographs using Deep Learning - Lattice the Machine Learning Journal Volume-2 Issue-2, ISSN 2582-8312

• LinkedIn Recommendations Hyperlink

• Awarded star employee of the month, Telerad Tech Pvt Ltd. Bengaluru
Received the award for the following : Published two articles in international peer-reviewed journal and to be able to present the same in the Conferences and Conclaves of both Medical and Engineering fields. Demonstration of dedication in the completion of development tasks ahead of timelines. For showing motivation, self-learning skills, and research initiatives.

• Awarded Outstanding Overall Performer of the Department of Information Science and Engineering, Acharya Institute of Technology, for the class of 2018
I was awarded for my excellent academic performance throughout the four years of my Bachelor's degree, as well as for my active involvement in various technical and cultural events, where I volunteered and coordinated several events.