I hold a Master's degree in Industrial and Systems Engineering from Rutgers University. On campus, I worked on time series analysis research for renewable energy systems and served as the operations lead at the solar electric-vehicle club.
I’m interested in working at companies that solve complex problems and tackle frontier markets, especially those involving high technical or scientific challenges. My goal is to leverage my engineering background and analytical skills to drive innovation and create impactful solutions in these dynamic industries.

My current projects include:

• (New!) An article on Fraud Detection in Supply Chains Using Kolmogorov Arnold Networks

• An article on how FinGPT can be revolutionary in the Indian ecosystem

Arena simulation was employed to optimize the queuing system at Rutgers University's on-campus Starbucks. Post-optimization, with additional resources for order processing, wait times dropped from 4.44 to 0.08 minutes for beverages. The analysis involved lognormal inter-arrival times, with R programming used for further distribution verification.

May 2023
Arena  •   Report

This project optimizes Mareana Turbine's production lines, identifying bottlenecks at QA and CNC stations through Flexsim simulation. Proposed enhancements include additional workstations for key impeller lines and adopting Industry 4.0 technologies like predictive maintenance and AI inspection to boost efficiency. The strategy aims to meet demand, increase revenue by $700K, and incorporate digital advancements at a $150K cost.

December 2022
FlexSim  •   Report

This project applies NLP and ML techniques to analyze Rutgers and UPenn subreddit discussions, performing feature engineering,employing multiple classifiers and then utilizing grid search on the best model for parameter tuning, then comparing the model custom Neural Network with Gemini API embeddings to discern thematic and sentiment trends. Results, visualized through t-SNE and word clouds.

May 2024
Github

Developed a dynamic chat bot using OpenAI's APIs and Langchain's OLLAMA Local for real-time language model interactions. Aimed to showcase seamless integration and operational efficiency, this project emphasizes the use of AI in enhancing conversational user experiences. It highlights the innovative combination of advanced language processing and local hosting capabilities, providing a robust framework for AI-driven chat applications.

April 2024
Github

Developed an application that utilizes Google's Gemini API and FAISS for efficient information retrieval from PDFs through conversation. Aimed to demonstrate vector database utility and intergrate it into Large Language Model interactions, offering users a responsive, context-aware chat experience with their documents. The project underscores the blend of AI and database search advancements for real-world applications.

March 2024
Github

Embarked on an NLP quest with Word2Vec to unlock semantic relationships in the Harry Potter series. In this project we explore into word embeddings, using the rich narrative to model linguistic contexts and uncover patterns. This technical exploration sheds light on character interconnections and thematic elements, providing a playground for AI-driven literary analysis.

March 2024
Github

Implemented a Convolutional Neural Network to classify handwritten digits from the MNIST dataset. The goal is to achieve over 90% accuracy, and the provided example reaches 98.57% accuracy. Covers data preprocessing, model architecture explanation, training parameters, and evaluation steps.

September 2023
Colab

This model uses dense layers to encode images into latent variables, employs reparameterization for sampling, and decodes to reconstruct images. Training focuses on optimizing latent dimensions to 2 and minimizing reconstruction loss through binary cross-entropy. Post-training, it quantitatively analyzes latent space (µ, σ) for each digit.

November 2023
Colab

The code generates n random sequences of observations (each of size 20 or higher). Each random sequence is decoded to find the best set of observations to describe it. If the system is not at high risk, the resulting sequences of states is labeled 'Good' and vice versa. If the number of bad labels in the sample of n is greater than a threshold (≥ 0.3n), then a control is applied with ∆= 0.2 and the steps above are repeated.

October 2023
Colab

The model sequentially processes 62 timesteps of features, including position, velocity, and acceleration, to predict future coordinates. An LSTM layer with 50 units is followed by two dense layers with 30 and 10 units, all employing ReLU activation, culminating in an output layer that forecasts the vehicle's position for the next 5 timesteps. Trained on a dataset of 18,457 files over 60 epochs with a batch size of 32, the model was optimized using the Adam optimizer and Mean Squared Error (MSE) as the loss function.

September 2023
Colab

The environment consists of a 10x10x10 grid with 15 randomly placed obstacles, penalties for staying in place (-2), collisions (-2), a step penalty (-1), and a reward for reaching the goal (10). The Q-learning approach, with parameters alpha = 0.1, gamma = 0.99, and epsilon = 0.1, iterates over 10,000 episodes, adjusting actions to maximize rewards. Post-training, the algorithm determines an optimal policy, demonstrated through an animation of the agent's path, highlighting the model's efficiency in navigating complex environments.

November 2023
Colab

Employing Monte Carlo simulations and savings growth against investment strategies of varied SPY/TLT allocations over 5, 7, and 10 years, this report provides a quantitative pathway for Jane Doe to accumulate a $500k down payment for a $1.5 million condo. A 70/30 SPY/TLT allocation shows the highest 10-year mean capital at $977,952.42 post-tax, inclusive of a 7% salary growth.

December 2023
Colab  •   Report

Rutgers University's Lot 64 Solar Canopy Project, at an investment of $1,623,116, aims for sustainable energy generation and EV charging infrastructure. The economic analysis includes a 7.42-year break-even point, based on annual revenues of $291,000 from solar savings and $80,000 from EV charging, offset by $72,000 in maintenance costs. The project exhibits a positive NPV of $366,140.20 and an IRR of 12.84%, surpassing the 10% discount rate, indicating robust financial health.

December 2023
Colab  •   Report

This technical report conducts a quantitative risk assessment of lidar system failures in autonomous vehicles using fault tree and event tree analyses, reliability and uncertainty analysis, and monetary risk assessment. Key findings include a lidar system failure probability of 11.66%, with laser emitter failure identified as a critical failure point. Reliability modeling, based on the Weibull distribution, estimates a laser emitter mean time to failure (MTTF) of 8.399 years.

December 2022
Report

This is the first part of the news dataset analysis just to know what sector to look into for a deeper dive into the dataset.

Feburary 2024
Colab

Utilize an LSTM network for forecasting PGE NP15's Day-Ahead Locational Marginal Pricing (DA_LMP), analyzing over 26,000 hourly data points from California's electricity load and gas prices for 2020 and 2023. Employing a Sequential LSTM model trained to minimize RMSE, the study achieves forecasts primarily within the $150 to $175 per MWh range. The model’s performance, characterized by its ability to identify price spikes, underscores the LSTM's effectiveness in navigating the complexities of energy market pricing.

March 2024
Colab

Using datasets on team and player performance, injuries, and managerial tenure, I applied models like linear regression, stepwise regression, and principal component regression while displaying the findings using Tableu. Key findings include Manchester City predicted as the season's top team, Mohamed Salah as the top scorer, and evidence confirming home advantage in football.

December 2022
R-Studio  •   Report