Nikita Dalvi | Data Scientist

Hi, I'm Nikita Dalvi.

A

Dedicated to the field of data science and machine learning, I am committed to continuous learning and adept at deriving actionable business insights.

About

I am a recent graduate from the University of Southern California, where I majored in Machine Learning and Data Science. Currently, I work as a Data Analyst at USC, leveraging BI solutions and ArcGIS mapping to analyze patrol patterns and improve crime prevention strategies. My professional journey includes 2.5 years at Baker Hughes and an internship at Semio, where I honed my skills in data analysis and machine learning. Passionate about solving complex business problems, I combine mathematics and programming to deliver impactful insights. Additionally, I have expertise in Natural Language Processing and am exploring innovations in fine-tuning pre-trained LLMs.

Languages: Python, C++, Java, JavaScript
Databases: MySQL, PostgreSQL, MongoDB
Libraries: NumPy, Pandas, Scikit-Learn, Scipy, Matplotlib, Seaborn, OpenCV
Frameworks: PyTorch, Keras, TensorFlow, Express
Tools & Technologies: Power BI, ArcGIS, Git, Docker, AWS, GCP, Azure, Heroku

Looking for an opportunity to work in a challenging position combining my skills in Machine Learning and Data Science, which provides professional development, interesting experiences and personal growth.

Experience

University of Southern California

Data Analyst

Design and implement data models and architectures for crime analysis
Utilize statistical languages like Python and SQL for data extraction and analysis
Work with relational and non-relational database systems to manage data
Perform geospatial analysis of crime locations and patrol patterns to optimize resource allocation
Develop interactive dashboards and visualizations using Power BI and ArcGIS
Enhance communication, critical thinking, and organizational skills to support decision-making
Tools: Microsoft Power BI, Esri ArcGIS, Python, Javascript, MS Excel

August 2024 - Present | Los Angeles, CA

Semio.ai

Machine Learning Research Engineer

Developed a UNet based Transformer architecture to reconstruct lost part of image using GenAI and collaborated with CV team to integrate YOLO v5 based object detection system yielding accuracy up to 89% against baseline model achieving 72% accuracy
Deployed trained image inpainting and object detection model leveraging Google Cloud's Vertex AI Platform, ensuring seamless deployment and scalability
Tools: Python, PyTorch, GCP, Computer Vision

June 2023 - August 2023 | Los Angeles, CA

Baker Hughes

Graduate Engineer Trainee

Raised laboratory revenue from $450k to $1M by calibrating and installing Flow, Moisture, and Gas Sensors
Built real-time data pipeline to collect data from sensors. Utilized collected data for visualizing sensor performance using data analysis techniques, and developed Machine Learning models predicting lifetime of sensors
Assisted in laboratory and inventory management audits to create efficient supply chain practices
Involved in Health, Safety and Environmental Practices for manufacturing plant
Tools: Python, C++, Shell Scripting, SAP

July 2018 - August 2020 | Pune, India

Bharat Electronics Limited

Software Developer Intern

Built standalone application for Surveillance using Morphological operations, Filtering, Color detection and Object detection etc. using MATLAB and Raspberry Pi. Achieved ambiguous activity detection with accuracy of 84 %.
Publication: Published in IEEE 4th International Conference on Computing, Communications, Automation and Control (2018).
Tools: OpenCV, MATLAB

October 2017 - March 2018 | Pune, India

Projects

                
Music Recommendation System
                  Content based and collaborative filtering for recommedation.
                
AccomplishmentsCollected data using spotify API
Cleaned and preprocessed data
Implemented hybrid recommedation approach using content based and collaborative filtering

Flight Fare Predictor
                  Flight fare data analysis and predictive modeling.
                
AccomplishmentsAnalyzed flight fare trends using statistical analysis techniques
Developed and optimized ML model using Logistics Regression, and ensemble methods to accurately forecast trends, leveraging a large dataset comprising 300K flights
Achieved 0.15 Mean Absolute Error (MAE) with XGBoost Regressor, surpassing baseline model with 0.32 MAE
Executed seamless live deployment by integrating with GitHub repository through AWS CodePipeline, facilitating streamlined updates and maintenance on AWS Elastic Beanstalk

PET-PALS: A socializing app for pet parents
                  A React and Node.js based web application.
                
AccomplishmentsCreated and tested website utilizing React and Node.js, fostering a community for pet owners to engage and collaborate
Leveraged AWS S3 for seamless data storage, and employed Lambda functions for robust REST APIs
Enhanced the OpenAI GPT-2 model for insightful pet-related query responses and seamlessly integrated it into the website using AWS SageMaker and API Gateway
Containerized the application using docker and deployed on AWS EC2 for continuos deployment and integration

Analyzing Customer Churn using Power BI
                  An extensive analysis of a Telecommunication company's customer dataset.
                
AccomplishmentsCleaned dataset & explored various characteristics of dataset to find correlations between them
Presented final dashboards conveying portable insights

Credit Card Risk Evaluation
                  Credit card risk analysis using Supervised Machine Learning.
                
AccomplishmentsTools: Python, Pandas, Seaborn, Predictive ML
Performed exploratpry data analysis
Applied dimensionality reduction and class imbalance handling techniques
Predictive 2-class ML algorithms to forecast possible default clients
Metrics comparison of predictive algorithms/li>

American Sign Language Recognition
                  Convolutional Neural Network Design to recognize ASL
                
AccomplishmentsTools: Python, PyTorch, Computer Vision
Designed different neural network architectures
Performed transfer learning
Metric comparison amongst all CNNs

TrojanMap
                  A simple C++ based map application.
                
AccomplishmentsTools: C++, SQL, Bazel
Similar functionality as google maps using Tree, Graph, Search and sort algortihms
Computational complexity analysis

Online Learning Improvement for Ads Recommender
                  Leveraging preference data to Improve online learning phase in a Multi-Armed Bandit Setting.
                
AccomplishmentsCharacterized quality of demonstration data to generate portable insights
Presentd a Warm Thompson Sampling Algorithm generating higher cumulative reward

Video Summarizer
                  A Multimedia processing based model that yields short summary of the given input video.
                
AccomplishmentsUsed Mean Absolute Deviation & Shannon Entropy to detect shots and subshots from a video
Presented an User Interface with multimedia controls using PyQT