Hi, I'm Sashvad (Sachin) Satishkumar
Computer Science + Data Science Student
A third-year undergraduate computer science student at University of Maryland, College Park, MD.
Contact meAbout Me
My introduction
Currently a student at the University of Maryland, passionately working in the fields of machine learning, pathology, and software engineering.
GPA
within AI/ML sectors
Skills
Programming Languages
C
Java
JavaScript
Python
TypeScript
Frameworks/Libraries
Flask
Keras
React.js
TensorFlow
Developer Tools/Technologies
Git
MongoDB
MySQL
NextJs
NodeJs
Tailwind
MySQL
Education
My personal journeyHigh School
Thomas Jefferson High Schoolfor Science and Technology
College
University Of MarylandCollege Park
Experience
Machine Learning Quant Developer
Sentinel Capital - Philadelphia, PA
- Engineered an Expert Advisor (EA) in MQL4/MQL5, integrating the A-Gimat Reversal (AGR) indicator to enhance automated forex trading decisions, improving entry and exit timing, reducing execution latency by 35% .
- Developed a Python-MetaTrader 5 pipeline using ZeroMQ, leveraging machine learning models and the Buy Sell Magic (BSM) indicator to confirm trend reversals, achieving a 30% increase in trade prediction accuracy
Junior Quantitative Analyst (Software)
University of Maryland - College Park, MD
- Pioneered an automated system that harnessed NLP models (BERT, GPT-4, Llama 3) to extract and analyze quantitative strategies from research papers, resulting in a 92% relevance scoring accuracy.
- Streamlined the trading analysts’ workflow by implementing optimized preprocessing techniques and harnessing Python-based AI frameworks, cutting analysis time by 70%.
Geospatial Science Research Intern
George Mason University - Department of Geospatial Information Sciences
- Spearheaded a research project that utilized geospatial data to analyze the impact of climate change on urban development, resulting in a 40% increase in data accuracy and a 25% reduction in data processing time.
- Developed a Python-based geospatial data visualization tool that integrated real-time satellite imagery and climate data, enhancing data interpretation and analysis by 50%
Bioinformatics/Pathology Research Intern
Dartmouth-Hitchcock - Emerging Diagnostic and Investigative Technologies
- Conducted a study on the application of deep learning models in the analysis of histopathological images, achieving a 90% accuracy rate in cancer detection and staging.
- Developed a Python-based image processing pipeline that integrated convolutional neural networks and transfer learning techniques to enhance diagnostic precision by 80%.
Utilized Python, R, and MATLAB to advance tumor identification techniques by developing and implementing novel algorithms, leading to a 30% increase in diagnostic accuracy.
- Leveraged TensorFlow/PyTorch frameworks to analyze 12+ cancer subtypes, enhancing diagnostic precision by 85%.
Projects
Most recent work
Detecting Deception using Microexpressions and Audio Sentiment Extraction: A Machine Learning Approach
The project involved the development of a cutting-edge, cost-effective lie detection system utilizing multimodal, non-intrusive methodologies to accurately identify deception. By leveraging convolutional neural networks (CNN) and natural language processing (NLP) technologies, along with additional machine learning frameworks, the system was designed to analyze microexpressions and audio sentiments for reliable deception detection. This innovative solution introduced a straightforward and economical approach to lie detection, significantly improving both accuracy and usability in real-world applications through the application of advanced machine learning techniques.
Publication
Human Mobility-Based Synthetic Social Network Generation
This project focused on developing agent-based mobility simulations that integrated dynamically evolving social networks to realistically model human interactions and mobility patterns. Using foot-traffic data from SafeGraph for Fairfax County, VA, the system allowed agents to form social connections based on shared visits to points of interest. The evolving network closely mimicked real-world social structures, outperforming traditional network generators like Erdos-Renyi, Watts-Strogatz, and Barabasi-Albert models in terms of clustering and realism, enhancing simulation accuracy by 60%.
Publication Research Paper
ArcticAI: Development of MOHS 3D Laboratory Automation
The ArcticAI project centered on developing a 3D laboratory automation system for MOHS surgery, significantly improving the efficiency of real-time tumor tissue resection by 150%. The team used advanced 3D construction libraries such as OpenMVG and Neural Recon to convert 2D images into detailed 3D models. The workflow involved image segmentation, inking, and contouring techniques to generate accurate tissue models. The system included a web application developed using Flask and THREE.js, allowing users to view and interact with the 3D models, enhancing both visualization and diagnostic precision.
GitHub Repository Project Synopsis
Omics Deep Ordinal Regression Staging Models
Achieved an 86% accuracy rate in cancer stage prediction by implementing advanced ordinal regression modeling across multiple cancer subtypes, utilizing sparse neural network layers for efficient predictor constraint. Increased efficiency by 75% and diagnostic precision to 89% by leveraging convolutional neural networks to model intricate disease pathways, outperforming traditional methods in accuracy and time-efficiency
GitHub Repository Project Synopsis
TACS: A Calibrated Highway Surveillance Dataset for Traffic Analysis
This project introduces the TACS dataset, which is designed to aid in the development of traffic monitoring algorithms using realistic highway surveillance video streams. By leveraging live video from the Virginia Department of Transportation, the dataset provides accurate annotations for vehicle positions, speeds, lane placements, and road congestion metrics. The study's key contribution includes a novel method of camera calibration that converts video streams to a top-down view, enabling precise vehicle tracking and real-world data extraction. With between 10,000 and 15,000 frames from each of the 10 surveillance cameras, TACS addresses gaps in existing traffic datasets by focusing on practical CCTV video quality and automating data extraction. The goal of the dataset is to simplify the preprocessing work for researchers, enabling them to focus on refining traffic analysis algorithms for applications such as speed estimation, accident prediction, and congestion analysis.
Publication
Detecting Deception using Microexpressions and Audio Sentiment Extraction: A Machine Learning Approach
The project involved the development of a cutting-edge, cost-effective lie detection system utilizing multimodal, non-intrusive methodologies to accurately identify deception. By leveraging convolutional neural networks (CNN) and natural language processing (NLP) technologies, along with additional machine learning frameworks, the system was designed to analyze microexpressions and audio sentiments for reliable deception detection. This innovative solution introduced a straightforward and economical approach to lie detection, significantly improving both accuracy and usability in real-world applications through the application of advanced machine learning techniques.
Publication
Human Mobility-Based Synthetic Social Network Generation
This project focused on developing agent-based mobility simulations that integrated dynamically evolving social networks to realistically model human interactions and mobility patterns. Using foot-traffic data from SafeGraph for Fairfax County, VA, the system allowed agents to form social connections based on shared visits to points of interest. The evolving network closely mimicked real-world social structures, outperforming traditional network generators like Erdos-Renyi, Watts-Strogatz, and Barabasi-Albert models in terms of clustering and realism, enhancing simulation accuracy by 60%.
Publication Research Paper