Open to full-time & working student roles

Nimesh
Bhavsar

I build 

Decision Scientist turned Data Science grad student. I build enterprise AI systems — RAG pipelines, agentic workflows, and predictive models — that ship to production and actually work.

Get in touch GitHub LinkedIn
$0
AWS cost savings delivered
0%
Manual processing reduced
0
Housing records analyzed
0
Open source projects

About me

Turning noisy data
into clear decisions

I'm Nimesh Jitendra Bhavsar, an Indian data scientist currently pursuing my Master's in Data Science at Technical University Dortmund, Germany.

Before grad school I spent a year at MuSigma Inc. as a Decision Scientist, building production-grade AI systems — from enterprise RAG pipelines for Fortune-50 clients to AWS cost-optimization algorithms that saved $1.8M annually.

I'm drawn to the intersection of statistical rigor and practical AI — Bayesian methods, interpretable ML, and systems that actually deploy.

Outside of work, I led NASA's HERC team at VIT, winning 1st place internationally. I also built a web scraper to land my apartment in Dortmund. Priorities.

🤖
Enterprise AI
RAG systems, agentic workflows, and LLM orchestration for production.
📐
Bayesian Modeling
Hierarchical models, posterior inference, and probabilistic reasoning.
🌀
Dynamical Systems
Data-driven control — SINDy, MPC, and reinforcement learning.
☁️
Cloud & MLOps
AWS infrastructure, storage optimization, and ML pipeline deployment.

Work Experience

Where I've worked

MuSigma Inc.
Jul 2024 – Jul 2025
Decision Scientist · Bengaluru, India
  • Developed Enterprise RAG systems for a Fortune-50 client — advanced retrieval optimization, semantic search, and intelligent context orchestration.
  • Built Agentic AI workflows for automated document parsing with specialized LLM agents, reducing manual processing effort by 15%.
  • Ran A/B tests on embedding models and context windowing strategies to reduce hallucinations in enterprise-scale LLM applications.
  • Designed a Predictive Customer Churn Model using PyTorch and XGBoost, outperforming existing benchmarks via advanced feature engineering.
  • Optimized AWS cloud storage via tiering algorithms, delivering a validated projected annual saving of $1.8 Million.
RAGLLM AgentsPyTorchXGBoostAWSA/B Testing

Education

Academic background

Technical University Dortmund
Current Oct 2025 – Present
M.Sc. Data Science · Grade 2.1/5 · 32 credits completed · Dortmund, Germany
  • Coursework: Applied Bayesian Data Analysis, Monte Carlo Simulation, Industrial Data Science, Deep Learning, Data Science for Dynamic Systems
Bayesian StatisticsDeep LearningDynamical Systems
Vellore Institute of Technology
Sep 2020 – May 2024
B.Tech Computer Science · CGPA 8.00/10 · Vellore, India
  • 🏆 Special Achiever Award — 1st Place in an International Competition organized by NASA.
  • 🚀 President, SAE Team Infinix — Official NASA Human Exploration Rover Challenge (HERC) Team.
  • Coursework: Database Management, Programming for Data Science, NLP, Information Security, Blockchain, AI, Social and Information Networks.
NASA HERCLeadershipComputer Science

Projects

Things I've built

🏠
Neighborhood Heterogeneity in Housing Prices
Bayesian Multilevel Modeling
Analyzed 21,613 housing transactions across 70 zip codes. Implemented multilevel Gaussian models in R (brms) with LOO-CV evaluation; varying-slope models best capture local market heterogeneity.
RbrmsBayesianStan
🌀
Learning Unknown Dynamical Systems via SINDy
Sparse Identification of Nonlinear Dynamics
Identified governing ODEs of an unknown 2D nonlinear system from 200 noisy trajectories using SINDy + LASSO. Recovered inward-spiraling dynamics with MSE 2.9×10⁻⁴ over a 100-step horizon.
PythonSINDyLASSOODE
🎯
Cart-Pole Swing-Up Control
MPC · SAC · Neural Network Surrogate
Unified comparison of Nonlinear MPC, Soft Actor-Critic RL, and MPC with PyTorch MLP surrogate dynamics. Evaluated on upright-time ratio, control effort, and max cart displacement.
PythonPyTorchRLMPC
🏡
Studierendenwerk Dortmund Freie Zimmer
Automation · Web Scraper
Web scraper that monitors available student accommodation on Studierendenwerk Dortmund's site, extracts room details and supports notification alerts. Ended up getting my apartment through it.
PythonScrapingAutomation
📊
RealGames Simulation Parameter Analysis
Agent-Based Market Simulation
Parameter sensitivity framework for an agent-based beverage market sim across 67 runs. Used Random Forest, XGBoost, SHAP, and ALE plots to identify drivers of market share, revenue, and adoption.
PythonXGBoostSHAPABM

Skills

Technical toolkit

Programming Languages
Python95%
R85%
SQL80%
Java65%
C++55%
Machine Learning & AI
RAG & LLM Systems90%
PyTorch / Deep Learning80%
XGBoost / scikit-learn88%
NLP / Embeddings75%
Statistical Methods
Bayesian Hierarchical Modeling GLM / Regression Hypothesis Testing Time Series Analysis Stochastic Processes Monte Carlo Simulation MCMC Diagnostics LOO Cross-Validation
Data Libraries
pandasNumPySciPy statsmodelsbrmsrstan tidyverseggplot2
Tools & Visualization
Power BITableauMatplotlib SeabornPlotlyAWS Git / GitHubLaTeX
Languages
English · Fluent Hindi · Native German · A1

Contact

Let's build something together

I'm actively looking for full-time positions and working student roles in data science, ML engineering, and AI research. Based in Dortmund — open to relocation.

✉️ Email nimeshbhavsar006@gmail.com 💼 LinkedIn linkedin.com/in/nimesh-bh ⚡ GitHub github.com/NimeshBhavsar
Send an email Message on LinkedIn