Data Scientist · Germany
I build production ML systems that are honest about their own failures.
Reviewing contracts is a chore. This makes it less of one. Ask a question in plain English, get a cited answer in seconds, without reading 50 pages yourself.
- LangGraph ReAct agent with four tools: semantic search, clause extraction, risk flagging, cross-contract comparison
- Fully local LLM via Ollama (Mistral 7B). No API keys, no data leaves the machine
- FAISS vector store with sentence-transformers embeddings
- Every answer cites the exact contract and chunk it came from
- Interactive UI built with Streamlit
A retrained model that quietly got worse is harder to catch than one that obviously broke. ML Guardian catches it before it reaches production. It compares every new model against the live one and blocks deployment if something regressed.
- Checks accuracy, F1, ROC-AUC, calibration, and feature drift against the current production model
- Configurable thresholds per use case
- Auto-promotes on pass, hard blocks with a full report on failure
- GitHub Actions native, no extra infrastructure needed
Built on top of diagnost and ml-production-template.
Most ML tutorials end at model.fit(). This one doesn't. It covers everything from training pipeline to served API so the next project starts from a solid foundation.
| Layer | Tool |
|---|---|
| Experiment tracking | MLflow |
| Training pipeline | Prefect |
| Model serving | FastAPI |
| Containerisation | Docker + Docker Compose |
| CI/CD | GitHub Actions |
| Quality gates | ml-guardian |
diagnost · 
Accuracy on a test set is one number. It does not tell you whether your model is overconfident, whether your data has drifted, or whether it performs equally across groups. Diagnost asks those questions so you do not have to remember to.
| Pillar | What it covers |
|---|---|
| Performance | Classification, regression, clustering metrics |
| Calibration | Reliability curves, ECE, Brier score |
| Drift Detection | KS and Chi-Square tests per feature |
| Dataset Diagnostics | Class imbalance, missing data, outliers, correlation |
| Model Comparison | Side-by-side benchmarking with winner declaration |
pip install diagnostDimensionality Reduction and Feature Extraction for High-Dimensional Datasets
FAU Erlangen-Nürnberg · 2025
ML & Modeling
Python · scikit-learn · XGBoost · PyTorch · Hugging Face
LLM & RAG
LangChain · LangGraph · FAISS · sentence-transformers · Ollama
MLOps & Infrastructure
MLflow · Prefect · FastAPI · Streamlit · Docker · GitHub Actions
Evaluation & Diagnostics
diagnost · calibration · drift detection · ECE · model comparison
Data
pandas · numpy · SQL · AWS
Junior Data Scientist
LINO Consulting and Research · Munich · Nov 2025 – Present
Data Analytics and Reporting · Werkstudent
Loyalty Partner Solutions GmbH · Munich · Jul 2023 – Oct 2025
Artificial Intelligence Student Assistant
Pattern Recognition Lab, FAU Erlangen-Nürnberg · Aug – Dec 2023
Senior Operations Executive
Infosys Limited · Pune · Dec 2020 – Aug 2022
M.Sc. Data Science — Major: AI & Machine Learning
FAU Erlangen-Nürnberg · 2025
Diploma in Data Analytics and Econometrics
Pune University · 2021
Bachelor of Business and Computer Applications
MIT World Peace University · 2020