Vispi Nevile Karkaria

Machine Learning • Digital Twins • Time-Series • Optimization

Explore

As a Ph.D. student at Northwestern University, my work centers on designing and deploying machine learning algorithms for Digital Twins. I build models for time-series analysis, deep learning, language modeling, uncertainty quantification, and optimization—aimed at real-time decision support. A Digital Twin is a live, data-driven replica of a physical system that learns continuously and feeds back actionable insights to its real-world counterpart. My research advances this physical–digital synergy to enable reliable, fast, and interpretable control.

Education

Formal training and academic milestones.

Northwestern University, Evanston, USA
Ph.D., Machine Learning & Data Science Algorithms
  • Sep 2021 – May 2026
  • GPA: 3.88/4.00
  • Advisor: Prof. Wei Chen
Northwestern University, Evanston, USA
Graduate Certificate in Predictive Science & ML
  • Sep 2022 – Sep 2024
  • GPA: 4.00/4.00
College of Engineering, Pune (COEP), India
B.E. Mechanical Engineering
  • Jul 2018 – Jul 2021
  • GPA: 9.84/10.00

Projects

Selected projects spanning agentic AI, neural operators, BO, and predictive maintenance.

RAG-Based AI Pharmacist Assistant
2023–Present • Algorithms: RAG, LLMs, Tool-Use, Safety
  • End-to-end RAG assistant for medication counseling & interaction checks (HF encoder + FAISS + cross-encoder reranker): 92% top-3 recall, 87% citation-backed answers, p95 1.4s on a 1M-doc index.
  • Guardrails (PII filters, dose-range rules, red-team prompts) + abstention: −38% hallucinations, −61% wrong-dose suggestions; Docker microservice with MLflow tracking.
View on GitHub
Multi-Agent LLM System for Automated Optimization
2023–Present • Algorithms: GPT, Agentic AI, RAG, LMs
  • API where users submit optimization problems; system selects & applies solvers via RAG.
  • Two-stage pipeline: method selection then solver invocation for accurate solutions.
Attention-Based Spatio-Temporal Neural Operator
2024–Present • Federated Learning, Neural Operator, Attention
  • +25% neural-operator accuracy via temporal–spatial attention.
  • Federated learning variant for secure cross-application transfer.
View on GitHub
RL-Transformer Tire Resource Allocation
2023–Present • Transformers, RL, Interpretable ML
  • +50% real-time tire lifespan prediction via Temporal Fusion Transformer.
  • RL policy extended tire health up to 95% with transformer-guided actions.
Bayesian Optimization for Time-Series (BOTS)
2022–2024 • Deep Learning, Transformers, BO
  • Transformer surrogate reduced AM prediction time; quantified predictive UQ.
  • Time-series optimization via BO on sequence profiles.
View on GitHub
Latent-Variable Constrained Bayesian Optimization (LV-CBO)
2022–2023 • Mixed Variables, GPs, PyTorch
  • Handled continuous + categorical via latent variables.
  • +20% operational efficiency, −15% cost on partner use cases.
Tire Lifespan Prediction via Random Forest
2021–2023 • RF, GP Models, Data Balancing
  • Robust pipeline for tire lifespan; VR-SMOTE outperformed standard balancing.
LSTM for Streamlining AM
2022–2023 • LSTM, Deep Learning, Big Data
  • LSTM surrogate significantly reduced prediction latency for AM.
  • Built big-data infra to manage high-volume process data.
ML to Enhance Super-Capacitor Lifespan
2020–2023 • DL, GPs, PyTorch
  • Predicted current output across configurations using DL.
  • Gradient-based search identified efficient designs.

Publications

Peer-reviewed journal articles and book chapters.

An optimization-centric review on integrating artificial intelligence and digital twin technologies in manufacturing
Karkaria, V.; Tsai, Y-K.; Chen, Y-P.; Chen, W. • Engineering Optimization, 2024
Paper Link
Towards a Digital Twin Framework in Additive Manufacturing: Machine Learning and Bayesian Optimization for Time Series Process Optimization
Karkaria, V. et al. • Journal of Manufacturing Systems, 2024
Paper Link
A Digital Twin Framework Utilizing Machine Learning for Robust Predictive Maintenance: Enhancing Tire Health Monitoring
Karkaria, V. et al. • arXiv:2408.06220, 2024
Paper Link
A Machine Learning–Based Tire Life Prediction Framework for Increasing Life of Commercial Vehicle Tires
Karkaria, V. et al. • Journal of Mechanical Design, 146(2), 2024
Paper Link
Tire Life Assessment for Increasing Re-Manufacturing of Commercial Vehicle Tires
Karkaria, V. et al. • Wiley, 2024, pp. 599–612
Paper Link
Digital twins for the designs of systems: a perspective
van Beek, A.; Karkaria, V.; Chen, W. • Structural and Multidisciplinary Optimization, 66(3), 2023
Paper Link

Conference Papers

Selected conference publications and proceedings.

"A Computational Framework for Social Entrepreneurs to Determine Policies for Sustainable Development."
Karkaria, V., et al. • ASME IDETC/DAC, 2021
https://doi.org/10.1115/DETC2021-70827
"Investigation of substitute jar materials for Laboratory-grade ball milling machine to process electrode materials for energy storage devices"
Shinde, S., Momin, T., Karkaria, V. N., Karandikar, P. B. • IOP MSE, 1206, 012018
doi:10.1088/1757-899x/1206/1/012018
"Electrode electrolyte compatibility for superior performance of super-capacitor"
Godse, L. S., Karkaria, V. N., et al. • IEEE PETPES 2019
doi:10.1109/PETPES47060.2019.9003864
"Sizing the connectors for super-capacitors"
Karkaria, V. N., et al. • ICECIC, 2019
Kanchipuram, India • pp. 1–7
"E-rickshaw present past and future with reference to current transportation in India"
Kokate, V. L., Bankar, D. S., Holmukhe, R. M., Karkaria, V. N., Karandikar, P. B. • ICRIEECE
Vol. 01(01), pp. 1–15
"Review of technologies for making electric vehicles as main mode of transport"
Karkaria, V. N., et al. • IEEE ICACCT, 2018
doi:10.1109/ICACCT.2018.8529591
"Thermal dicky"
Karkaria, V. N., Korgaonkar, P. R., Karandikar, P. B. • Institute of Engineers, Pune, 2017
pp. 90–96
"Innovative methods of ball milling to grind activated carbon as an electrode material for enhancing the performance of ultracapacitor"
Godse, L. S., Karkaria, V., Karandikar, P. B., Kulkarni, N. R. • IEEE ICECDS, 2017
doi:10.1109/ICECDS.2017.8389807