Skip to content
View shruti976's full-sized avatar
🎯
Focusing
🎯
Focusing

Highlights

  • Pro

Block or report shruti976

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
shruti976/README.md

Shruti Singh

CS PhD Candidate · ML / AI Researcher · Data Scientist

Portfolio LinkedIn Google Scholar Profile Views Email


I build reproducible machine-learning systems for reinforcement learning, large language models, computer vision, vision-language systems, and applied data science.

I am a Computer Science PhD candidate at the University of Dayton, with research experience spanning Safe Reinforcement Learning, Vision-Language Models, AI Safety, Human-AI Systems, and industry experience at Walmart and Padagis LLC.

🌐 Personal Website: https://www.shrutisinghphd.com/


🔬 Current Focus

  • 🧠 LLM-guided Safe Reinforcement Learning
  • 🤖 Adversarial robustness in RL environments
  • 👁️ Vision-Language Models for construction safety
  • 🏗️ AI systems for construction safety analysis
  • 📊 Applied data science and statistical modeling
  • 🔍 Human-AI collaboration and explainability

📊 GitHub Stats


🚀 Selected Projects

ML / AI

Adversarial Resilience in RL Systems

Entropy-driven feature selection for adversarial robustness; 94–95% accuracy across Gym (LunarLander, BipedalWalker), outperforming KL-divergence and joint-entropy baselines. Published at KSE 2024.

🔗 Repository: adversarial-rl-resilience

SiteCortex: Construction Safety VLM

Modular, glass-box vision-language pipeline for automated construction-safety hazard detection — captioning, hazard proposal, multimodal reasoning, hybrid visual grounding, and rule-based reconciliation into an auditable safety report.

🔗 Repository: sitecortex-construction-safety-vlm

Gridworld Search & Planning Strategies

Classical search (DFS, BFS, UCS, A*) and reinforcement-learning planning (value iteration, policy iteration, Q-learning) with trajectory and policy visualizations.

🔗 Repository: gridworld-search-strategies

Data Science

Temporal Forecasting with LSTM

Deep learning for time-series prediction with advanced preprocessing, hyperparameter tuning, learning-rate scheduling, and early stopping — benchmarked against MLP, CNN, and CNN-LSTM (LSTM achieves the lowest error).

🔗 Repository: lstm-time-series-forecasting

Deep RL for Portfolio Allocation

PPO agent that allocates a multi-asset ETF portfolio using a differential-Sharpe reward, benchmarked out-of-sample against equal-weight, mean-variance, and buy-and-hold with leakage-free backtesting.

🔗 Repository: drl-portfolio-optimization

Credit-Card Default Prediction (Explainable ML)

Gradient-boosted default prediction on 30k accounts with imbalance-aware metrics, probability calibration, and SHAP explanations for auditable, per-applicant credit decisions.

🔗 Repository: credit-risk-default-prediction


📚 Publications

Enhancing Sustainability and Construction Safety Research in the Era of Artificial Intelligence

Access here ASME JESBC 2025

Systematic review of AI technologies and their impact on construction safety research.

Mouse Brain Cell Segmentation in Fluorescence Microscopy Images

Access here CISS 2025

Segmentation of brain cells in fluorescence microscopy using deep learning methods.

Virtual Yoga Instructor with Real-Time Feedback

Access here CISS 2025

Real-time pose estimation and feedback system for guided virtual yoga instruction.

Imposter Injection: Learning to Select Features in Reinforcement Learning

Access here KSE 2024

Entropy-driven feature selection approach for adversarial robustness in reinforcement learning environments.

Predicting an Optimal Medication/Prescription Regimen for Patient Discordant Chronic Comorbidities Using Multi-Output Models

Access here Information, 2024

Multi-output modeling approach for personalized medication recommendations in patients with comorbidities.

Multi-Output Career Prediction: Dataset, Method, and Benchmark Suite

Access here CISS 2023

Benchmark suite and multi-output methods for career trajectory prediction.


⚙️ Technical Skills

Languages

Python · R · SQL · NoSQL

ML / DL

PyTorch · TensorFlow · Keras · scikit-learn · Hugging Face

Reinforcement Learning

DDQN · OpenAI Gym · LunarLander · BipedalWalker

Vision / VLM

OpenCV · BLIP-2 · GroundingDINO

Data Science

Pandas · NumPy · Matplotlib · Statistical Analysis · Time-Series Modeling

Data Engineering

BigQuery · ETL Pipelines


🎓 Education

University of Dayton

PhD in Computer Science

Focus:

  • Safe Reinforcement Learning
  • Machine Learning
  • Human-AI Systems

MS in Computer Science
Graduated with 4.00 GPA


📫 Connect

📧 shruti.singh97.phd@gmail.com

💼 LinkedIn
https://www.linkedin.com/in/shruti-singh97/

🎓 Google Scholar
https://scholar.google.com/citations?user=t2OTFtkAAAAJ&hl=en

🌐 Website
https://www.shrutisinghphd.com/

Popular repositories Loading

  1. drl-portfolio-optimization drl-portfolio-optimization Public

    Deep reinforcement learning (PPO) for multi-asset portfolio allocation with a differential-Sharpe reward, benchmarked out-of-sample against equal-weight, mean-variance, and SPY.

    Python 1

  2. shruti976 shruti976 Public

  3. adversarial-rl-resilience adversarial-rl-resilience Public

    Entropy-driven feature selection for diagnosing robustness failures in RL agents under adversarial sensor perturbations (LunarLander, BipedalWalker).

    Jupyter Notebook

  4. sitecortex-construction-safety-vlm sitecortex-construction-safety-vlm Public

    Glass-box vision-language pipeline for construction-site safety analysis: interpretable hazard parsing, grounding, and stage-by-stage evaluation.

  5. gridworld-search-strategies gridworld-search-strategies Public

    Classical search (DFS/BFS/UCS/A*) and RL planning (value/policy iteration, Q-learning) with policy and trajectory visualizations in Gridworld.

    Jupyter Notebook

  6. lstm-time-series-forecasting lstm-time-series-forecasting Public

    LSTM next-hour temperature forecasting on the Jena Climate dataset, benchmarked against MLP, CNN, and CNN-LSTM (TensorFlow/Keras).

    Python