Vishesh Srivastava
Data Scientist & AI Engineer
Professional Summary
I'm an AI/ML Engineer and Data Scientist based in Dortmund, with an M.Sc. in Data Science from TU Dortmund. I build end-to-end ML systems, from sensor time-series classification (published, Robert Bosch) to LLM-powered applications with RAG pipelines, fine-tuning, and cloud deployment.
My ML stack includes PyTorch, scikit-learn, LangChain, HuggingFace, TabPFN, XGBoost, and Qdrant. I deploy with FastAPI, Docker, GitHub Actions CI/CD, and cloud platforms (AWS, Azure, DigitalOcean). I've co-authored two peer-reviewed papers — IEEE IJCNN 2024 and Journal of Composite Materials 2025.
My full-stack background (3.5 years, C# / ASP.NET / SQL Server across enterprise domains) means I write production-grade code, not just notebooks.
Skills & Expertise
AI & Machine Learning
Cloud & MLOps
Data & Vector Stores
Programming
Work Experience
AI Engineer (Intern)
AICU GmbH
Heilbronn, Germany
July 2025 - October 2025
- •Designed and shipped an end-to-end CSV data-quality workflow (merge, type inference, cleaning, visualization) using Python / Pandas with a FastAPI backend, Streamlit UI, and SQLite audit logging.
- •Integrated the Together AI API for LLM-assisted data cleaning; experimented with fine-tuned open-source models (DeepSeek-Coder, LLaMA via QLoRA) and chained agentic steps using LangChain.
- •Containerized the platform with Docker and deployed to Streamlit Cloud and an Azure VM via Docker Compose + NGINX reverse proxy.
- •Delivered production-ready validation, logging, and automation components in agile sprints; received a formal employer reference letter commending analytical thinking and creative problem-solving.
Master's Thesis Researcher
Robert Bosch GmbH
Blaichach, Germany
May 2024 - January 2025
- •Built an end-to-end multimodal defect-classification pipeline for porosity prediction in glass-fiber-reinforced polymer (GFRP) components, contributing to an ongoing Industry 4.0 / Digital Twins research programme.
- •Image branch: trained a multitask ResNet-50 CNN (R² = 0.85) and a 38-feature Elastic Net (R² = 0.93); ensembled both for image-based porosity estimation.
- •Time-series branch: benchmarked TabPFN, AutoGluon, and XGBoost on 129 features extracted from BMA456 MEMS accelerometer signals using Bayesian HPO and 10-fold CV — TabPFN achieved F1 = 0.80, AUC = 0.98 on a small (n = 323), imbalanced dataset.
- •Delivered full interpretability analysis (SHAP, calibration curves, ablation studies) and co-authored a peer-reviewed paper in Journal of Composite Materials (2025).
AI/ML Engineer & Scholarship Recipient
Hessen Ideen Stipendium
Remote, Germany
July 2023 - December 2023
- •Awarded a 6-month state scholarship by the Hessisches Ministerium für Wissenschaft und Forschung to build an agentic project-management automation system.
- •Built FastAPI services and document-processing pipelines that converted unstructured files into structured, validated outputs for downstream agentic workflows using LangChain and HuggingFace Transformers.
- •Containerized all services with Docker and participated in the Hessian AI programme for market validation of AI-driven PM tooling.
Working Student — C# Backend Developer
choyze GmbH
Hannover, Germany (Remote)
August 2023 - February 2024
- •Enhanced and maintained backend C# / .NET applications powering the HR-tech product suite; implemented new REST API endpoints and feature modules.
- •Optimized SQL Server queries, stored procedures, and triggers to improve latency on high-volume transactional flows.
Full-Stack Developer / System Analyst
EPAY Systems · Contata Solutions · Aon
India
July 2017 - December 2020
- •Delivered enterprise applications across HR, insurance, payroll, petroleum-refining, and logistics domains using C# / ASP.NET MVC, SQL Server, and JavaScript.
- •Designed relational schemas and optimized SQL Server stored procedures and triggers for high-volume transactional systems.
- •Owned the full software lifecycle across multiple product cycles: requirements, development, debugging, release, and production maintenance.
Publications
Autoencoder Optimization for Anomaly Detection: A Comparative Study with Shallow Algorithms
V. Kumar, V. Srivastava, S. Mahjabin, A. Pal, S. Klüttermann, E. Müller — IEEE International Joint Conference on Neural Networks (IJCNN), 2024
Innovative guide for optimizing autoencoder performance targeting anomaly detection. Explores latent space selection, optimal latent size determination, and diverse hyperparameter tuning across image and tabular datasets (ADBench), ensuring autoencoders match or surpass shallow baselines.
DOI: 10.1109/IJCNN60899.2024.10650057
Void Content Classification for Acceleration Sensor Embedded Glass-Fiber Reinforced Components Made by Resin Transfer Molding
M. Münch, V. Srivastava, P. Middendorf — Journal of Composite Materials, 2025
Local void content classification using embedded acceleration sensor signals during RTM production monitoring. Employed ML algorithms with nested cross-validation; CV estimates exceeded 89% for F1, Recall and Precision. Research conducted at Robert Bosch GmbH / Bosch Sensortec.
DOI: 10.1177/00219983251391884
Certifications & Courses
Education
Master of Science — Data Science
Technische Universität Dortmund
Dortmund, Germany
October 2021 - November 2025
GPA: 2.4 (German scale: 1.0 = best, 4.0 = pass)
Activities & Societies:
- Master's Thesis Researcher at Robert Bosch GmbH (Deep Learning / Industrial AI)
- AI Engineer Intern at AICU GmbH (GenAI, LLM-powered automation workflows)
- Hessen Ideen Scholarship Project — Agentic project-management automation with LLMs
Bachelor of Technology — Information Technology
Guru Ghasidas Vishwavidyalaya
Bilaspur, India
August 2013 - April 2017
GPA: 7.37 / 10 (First Division)