Projects & Publications

From AI research breakthroughs to MedTech innovations, here's how I'm building the future of technology.

15+ Projects

Across AI, Tech, and social impact

3 Publications

In topics I'm passionate about

5 Startups

Founded or contributed to

Featured Work

A selection of projects that showcase my passion for AI, MedTech, and social impact

Featured Projects

AI/ML

AI-Powered Urban Planning Tool

Completed

Agentic workflow system using IBM Watson models and cutting-edge AI to optimize urban planning decisions, resource allocation, and infrastructure development.

Key Achievements:

  • IBM Hackathon Winner
  • Multi-agent orchestration
  • Real-time urban analytics
IBM WatsonAgentic AIPythonGeospatial Analysis+1
AI/ML

Generative AI Customer Support

Production

Enterprise chatbot system built for Dell Technologies, serving 50K+ users with 60% reduction in ticket resolution time.

Key Achievements:

  • 50K+ users served
  • 60% faster resolution
  • Enterprise deployment
OpenAI APILangChainFastAPIReact+1

Other Projects

MedTech

Skin Disease and Solution Platform (SkinSnap)

Completed

Revolutionary diagnostic tool that uses computer vision and deep learning to analyze medical images, reducing diagnostic time by 40% while improving accuracy.

Key Achievements:

  • 10K+ images processed daily
  • 92% diagnostic accuracy
  • Conrad Challenge Award Winner
PythonTensorFlowOpenCVAWS+1
MedTech

Smart ECG Analysis System

Completed

Machine learning system for real-time ECG analysis and cardiac abnormality detection, developed during MIT Beaver Works program.

Key Achievements:

  • 92% accuracy rate
  • Real-time processing
  • MIT collaboration
PythonScikit-learnSignal ProcessingFlask
Social Impact

Violence Prevention Education Platform

Deployed

Interactive digital curriculum platform that reduced reported incidents by 30% across pilot schools in the CUHSD district.

Key Achievements:

  • 2000+ students reached
  • 30% incident reduction
  • 25 educators trained
ReactNode.jsMongoDBD3.js
AI Research

Autonomous Vision System

Published

Computer vision model for autonomous systems with 25% improved object detection accuracy using novel attention mechanisms.

Key Achievements:

  • 25% accuracy improvement
  • Conference publication
  • MIT collaboration
PyTorchOpenCVCUDAROS
AI/ML

Scalable Statistical Timing Engine (C++)

Completed

Built a high-performance C++ engine for timing analysis on large graph-based systems, incorporating probabilistic modeling to simulate variation and uncertainty. Designed incremental update algorithms to reduce re-computation by 80% on dynamic graphs.

Key Achievements:

  • 80% reduction in re-computation
  • 100k+ node graphs
  • Probabilistic modeling
C++Graph AlgorithmsStatistical ModelingPerformance Optimization
AI/ML

High-Performance C++ Graph Simulator

Completed

Implemented a C++ simulator for large-scale dynamic graphs (100k+ nodes), optimizing traversal and memory usage. Achieved 4× faster execution through incremental updates and profiling-driven optimizations.

Key Achievements:

  • 4× performance improvement
  • 100k+ nodes supported
  • Memory optimized
C++Data StructuresAlgorithmsMemory Management
AI Research

AI-Based Bias Analysis in Resume Screening

Published

Designed and evaluated algorithms to detect and mitigate bias in automated resume screening systems, using statistical analysis to improve fairness and reliability at scale.

Key Achievements:

  • Gender bias detection
  • Fairness metrics developed
  • Journal publication
AlgorithmsStatistical AnalysisMachine LearningEthics in AI
MedTech

Neural Network for Prenatal Anomaly Prediction

Completed

Built a deep learning pipeline to predict prenatal anomalies from medical imaging data, integrating uncertainty estimation to flag high-risk cases and improve model reliability.

Key Achievements:

  • High-risk case detection
  • Uncertainty quantification
  • Medical imaging analysis
PythonNeural NetworksStatistical ModelingData Pipelines
AI/ML

Interactive Graph Optimization Tool

Completed

Developed a graph-based optimization and visualization tool for large networks, enabling efficient traversal, real-time updates, and intuitive analysis. Recognized at multiple hackathons for innovation and performance.

Key Achievements:

  • Hackathon awards
  • Real-time updates
  • Large network support
Graph AlgorithmsOptimizationSystems ThinkingVisualization

Research Publications

Contributing to the academic community through peer-reviewed research in AI, ML, and ethics

Machine Learning-driven Resume Screening: Challenges and Implications for Gender Bias

Featured
Solo Author Publication
Ishita Samadhiya (Advised by Angelina Wang)
Horizon Inspires Academic Research Projects (HARP) Journal2024Academic Journal Paper

Machine Learning (ML) has revolutionized hiring processes, introducing new dynamics and challenges, especially in the realm of resume screening. This study delves into the challenges and implications of ML-driven resume screening, specifically addressing gender bias. We utilized a dataset containing resumes from different sectors in India to train models, including Random Forests and Multilayer Perceptron Classifier. These models were employed to categorize resumes into sectors such as technology, business, etc. Our results reveal gender biases, with men more frequently predicted as executives and women in technical roles, reflecting historical disparities. Furthermore, we identify limitations and ethical concerns surrounding such classifiers, emphasizing the need for responsible AI deployment in recruitment processes. By shedding light on the complexities of bias in ML-based recruitment, this research contributes to the ongoing discourse on ethical AI deployment, offering insights and recommendations for fostering fairness and accountability in hiring-based automated decision-making processes.

Using Deep Learning to Predict Mitochondrial, Multi-factorial, and Single-gene Genetic Disorders

Featured
2nd Award - Physical Science and Engineering Category
Ishita Samadhiya
Synopsys Silicon Valley Science and Technology Championship2024Conference Paper

In this era of abortion bans and restricted reproductive healthcare, we present an imbalance-aware deep learning framework for predicting mitochondrial, multifactorial, and single-gene genetic disorders from prenatal clinical metrics. The model uses a multi-encoder architecture with attention-based feature fusion and focal loss with adaptive class reweighting to address extreme rarity and heterogeneous inputs. The approach achieves 81% accuracy with an F1-score of 0.36, prioritizing clinically meaningful recall for early, non-invasive prenatal risk stratification while emphasizing fairness, uncertainty, and responsible deployment.