AI/ML Software Engineer @ UMD | MSML Student
I'm an AI Engineer and Machine Learning professional with 2+ years of experience in analyzing large datasets, building data visualization dashboards, and creating detailed reports for executive leadership. Currently pursuing a Master's in Machine Learning at the University of Maryland while working as an AI/ML Software Engineer.
My expertise spans multiple domains:
I specialize in leveraging cutting-edge technologies like Claude API, OpenAI, and advanced machine learning techniques to solve complex problems across various domains.
My journey in the world of data has equipped me with a diverse skill set, allowing me to contribute effectively across various roles in the AI and data ecosystem. I thrive on understanding the complete data lifecycle, from raw data ingestion to deploying intelligent applications.
Skills: Python, C++, TensorFlow/Keras, PyTorch, MLOps (MLflow), Docker, CI/CD, API Development (REST, FastAPI), AWS (SageMaker, CodeCatalyst, Lambda, EC2, S3), Microservices.
Skills: Python, XGBoost, LightGBM, Deep Learning (CNNs, RNNs, Transformers), NLP (Hugging Face, spaCy), Computer Vision (OpenCV), Model Evaluation & Tuning, Feature Engineering.
Skills: Python (Pandas, NumPy, SciPy), SQL, Statistical Modeling, A/B Testing, EDA, Data Visualization (Matplotlib, Seaborn, Plotly, PowerBI, Tableau), Jupyter Notebooks, Hypothesis Testing.
Skills: SQL, Python (Pandas, NumPy), R, Advanced Excel, Data Visualization (Tableau, PowerBI, Matplotlib, Seaborn), Statistical Analysis, A/B Testing, Report Generation, Data Cleaning, EDA (Exploratory Data Analysis).
My ability to seamlessly transition between these roles stems from a strong foundation in software engineering principles, a deep understanding of mathematical and statistical concepts, and hands-on experience across the entire AI/ML project lifecycle. I focus on:
Used Claude API and CatBoost to extract features and predict bill passability for the 2025 legislative session with 80% accuracy.
Created an NLP tool using GPT-3.5 and Langchain to analyze, comprehend, and summarize customer feedback with advanced sentiment analysis.
Developed a sophisticated financial analysis tool to analyze customer spending patterns and risk assessment for lending businesses.
Created a Bureau-based Income Estimation Model to verify income in digital lending journeys, enhancing financial risk assessment.
Trained a Variational Autoencoder to compress and predict future states in simulated autonomous vehicle environments.
Developed a cashflow forecasting architecture using multiple statistical and deep learning models, improving accuracy by 60%.
Validated foundational AWS Cloud knowledge and understanding of core AWS services, security, architecture, pricing, and support.
Validated foundational AWS Cloud knowledge and understanding of core AWS services, security, architecture, pricing, and support.
Comprehensive certification covering data analysis techniques, tools, and best practices using Google's professional methodologies.
Advanced certification by Andrew Ng covering neural networks, deep learning techniques, and machine learning project structuring.
Note: A comprehensive list of all my certifications can be found on my LinkedIn Certifications page.
Successfully organized a Data Science Hackathon to hire interns and full-time employees, demonstrating strong organizational and leadership skills.
Maintained a stellar academic record with a GPA of 4.0/4.0 in Master's program and 3.7/4.0 during Bachelor's degree.
This project aimed to predict the passability of legislative bills for the upcoming 2025 session. By leveraging advanced NLP techniques with the Claude API for feature extraction from bill texts and employing a CatBoost classifier, the model achieved an 80% accuracy in predicting whether a bill would pass or fail.
One major challenge was handling the diverse and often nuanced language of legislative documents. Fine-tuning the prompts for the Claude API was crucial for extracting meaningful and predictive features.
The model provides a valuable tool for lobbyists, policymakers, and analysts to forecast legislative outcomes with a high degree of accuracy, allowing for more strategic planning and resource allocation.
Developed an NLP-powered tool to automate the analysis and summarization of large volumes of customer feedback. This tool utilizes GPT-3.5 via OpenAI's API and the Langchain framework to perform sentiment analysis, topic modeling, and generate concise summaries.
This tool significantly reduces the manual effort required to process customer feedback, enabling businesses to quickly identify key issues, understand customer sentiment, and respond more effectively to customer needs.
Developed a sophisticated financial analysis tool to analyze customer spending patterns and risk assessment for lending businesses. The tool processes bank statements to extract transaction patterns, categorize spending, and generate risk scores for credit decisions.
Enables lending businesses to make data-driven credit decisions by automating the analysis of customer financial behavior and reducing manual review time.
Created a Bureau-based Income Estimation Model to verify income in digital lending journeys. The model leverages alternative data sources to estimate and validate borrower income when traditional verification is unavailable.
Enhances financial risk assessment by providing reliable income verification in digital-first lending environments, reducing fraud and improving approval accuracy.
Trained a Variational Autoencoder (VAE) to compress and predict future states in simulated autonomous vehicle environments. The world model learns a compact latent representation of the environment and uses it to predict future observations, enabling more efficient reinforcement learning.
Demonstrates the use of world models to improve sample efficiency in RL by learning predictive representations of the environment, applicable to autonomous driving and robotics.
Developed a cashflow forecasting architecture using multiple statistical and deep learning models to handle irregular time series data. The hybrid approach improved forecasting accuracy by 60% compared to baseline methods.
Provides robust cashflow predictions for businesses with irregular transaction patterns, supporting better financial planning and liquidity management.