M.Sc. Computer Science · Paderborn University · Germany

I'm a Master's student in Computer Science at Paderborn University, focusing on Data Science, Machine Learning, and privacy-preserving AI. My Master's thesis direction is Data Science.

I enjoy the part of ML where the math meets the system — turning a research idea into a working API, taking a paper from arXiv and reproducing its results, or finding a closed-form trick that replaces hours of training with a single matrix operation.

Currently looking for Werkstudent / working student roles in Data Science, ML, or applied AI across Germany.

• 🔬 Fast Unlearning on Linearized Face Recognition Models — year-long research project (Project Group, 20 ECTS) co-authoring a paper on a closed-form, SVD-based machine unlearning method for face recognition. Achieved Forget Scores of 0.97 / 0.99 on iResNet50 backbones (ArcFace / AdaFace) while degrading verification accuracy by only 0.45% / 0.05% across LFW, CFP-FP, AgeDB-30, CALFW, CPLFW. Lead contributor. Code in private research org repo; releasing on paper publication.

• 🎓 PaderBot — Multilingual RAG Q&A for International Students · Live Demo — Bilingual (English / German) RAG system over 94 pages from uni-paderborn.de and the Studierendenwerk (586 chunks). Hybrid retrieval combining multilingual e5-base dense embeddings with BM25, fused via Reciprocal Rank Fusion. Two-model Llama setup via Groq (8B for gated query rewriting, 70B for grounded generation), with confidence-gated refusal and source citations. Evaluated on a hand-built 30-question benchmark with from-scratch RAGAS-style metrics — 0.80 faithfulness, 0.91 answer relevance, with English/German faithfulness gap under 0.05.

• 🌱 Harvest Helper — Weather-Based Crop Recommendation API — FastAPI microservice within a dual-API agricultural advisory platform for North Rhine-Westphalia. Gradient Boosting Classifier on an 11-feature synthetic agronomic dataset (5,500 samples, 22 crops); 89.73% top-1 accuracy, 1.36 ms inference latency. Fuses real-time OpenWeatherMap data with DWD 1991–2020 climate normals. Co-authored IEEE-format paper. Module owner within a team of 4.

• 🕸️ - XAI Knowledge Graph — Research-paper knowledge graph (Neo4j + RDF/SPARQL/OWL + PyKEEN) with a GraphRAG pipeline benchmarked against vanilla RAG. RotatE precision@10 = 0.90; GraphRAG leads completeness 4.4 vs 2.5.

Languages       Python · SQL · C++ · Java
ML & Data       PyTorch · TensorFlow · scikit-learn · NumPy · Pandas · Matplotlib
LLM & RAG       Sentence-Transformers · multilingual-e5 · ChromaDB · BM25 ·
                Reciprocal Rank Fusion · Llama (via Groq, Paderborn AI-Gateway) ·
                prompt engineering · RAG architectures
ML Engineering  FastAPI · Streamlit · REST APIs · HuggingFace Spaces
Databases       MySQL · PostgreSQL · ChromaDB
Tools           Git · GitLab · Jupyter · Google Colab · VS Code 

• 🚗 Automatic Number Plate Recognition (ANPR) System — YOLO + OpenCV + OCR pipeline optimised for Indian license plates; 95% recognition accuracy across varied lighting conditions. (B.Tech. project)

• ✉️ Cold-Email-Generator — LLM-powered tool that turns job posts into personalised cold-outreach emails via a Streamlit interface.

• M.Sc. Computer Science, Paderborn University · 2025 – present · Current GPA 2.2 (German scale)
• B.Tech. Information Technology, GTBIT, GGSIPU · 2020 – 2024 · 9.28 / 10.0 CGPA

🇬🇧 English (C1) · 🇩🇪 German (A2) · 🇮🇳 Hindi (Native)

Open to Werkstudent opportunities across Germany · available immediately