CV

Applied AI & LLM Systems

Independent

• Built and open-sourced Sift: grounded question-answering over technical PDF corpora. Docling chunking, hybrid BM25 and dense (Chroma) retrieval, cross-encoder reranking, served over MCP to an agent that answers only from retrieved passages, with a source and page citation on every claim.
• Built an agentic tutoring system in daily use: an LLM agent orchestrating retrieval over a structured knowledge base, persistent memory of errors and progress, adaptive lesson selection that targets weak areas, and voice I/O (faster-whisper ASR, edge-tts TTS).
• Published technical writing on LLM internals and systems: attention mechanisms, data quality for LLM training, and the shift from generative to agentic AI.
• Deepened modern LLM architecture and inference-efficiency internals through Stanford coursework: KV-cache, speculative decoding, FlashAttention.

Senior AI Engineer

Flinn.ai

• Led development of LLM-powered product features for a platform serving medical device manufacturers, including data extraction from medical research papers and multilingual complaint monitoring.
• Partnered with product and backend teams to scope problems, define success metrics, and deliver features end to end.
• Navigated cost-performance tradeoffs under production constraints.

Career break

Parental Leave

Research Scientist

Amazon AI

• Led development of a scalable noise-removal/data-quality pipeline processing billions of tokens to improve training data for LLMs.
• Presented findings and recommendations to senior science and engineering leadership; mentored Master’s and Ph.D. interns on research deliverables.

Machine Learning Scientist

German Research Center for Artificial Intelligence (DFKI)

• Implemented transformer-based ML systems end-to-end (GPU training, Docker, production-level code) within BMBF-funded projects; published results at AAAI, NAACL, and EMNLP.
• Collaborated with academic and industry partners internationally; supervised interns and BSc/MSc students.

Machine Learning Scientist Intern

NVIDIA Research

• Built a document understanding pipeline combining table/cell detection, tabular structure retrieval, and OCR for financial documents.

Applied Scientist Intern

Amazon Alexa

• Developed a method for generating diverse synthetic training data to improve intent classification and slot labeling for task-oriented NLU.

Research Intern / Master's Thesis

SAP AI Research

• Designed and implemented an evaluation metric for Visual Question Answering (VQA) models.

Ph.D.

TU Kaiserslautern

Computer Science

• Grade: Sehr Gut 1.0 (highest on 1.0–5.0 scale)
• Thesis: Building Natural Language Generation and Understanding Systems in Data-Constrained Settings

Visiting Researcher

University of Copenhagen

Natural Language Processing

• Developed an unsupervised post-editing algorithm to generate fluent fact-checking explanations

M.Sc.

TU Kaiserslautern

Computer Science — Minor in Economics

• Grade: Sehr Gut 1.5

Semester Abroad

Kyushu University

Computer Science

• Explainable AI to analyze behavior of deep CNN architectures for image recognition

B.Tech

Guru Nanak Dev Engineering College

Computer Science & Engineering

• Grade: First Division with Distinction

Search and Learn: Improving Semantic Coverage for Data-to-Text Generation

EaSe: A Diagnostic Tool for VQA Based on Answer Diversity

Can Pre-training Help VQA with Lexical Variations?

Data-Efficient Paraphrase Generation to Bootstrap Intent Classification and Slot Labeling for New Features in Task-Oriented Dialog Systems

How Do Convolutional Neural Networks Learn Design?

AAAI-22 Scholarship by Hitachi

AI Newcomer 2021 Award

• Recognized by the German Informatics Society and BMBF, Germany.

EU-Cost STSM Grant

• Awarded by EU-Cost Action to work on the Multi3generation project at CopeNLU, University of Copenhagen.

Best Student Paper Award

• Awarded by the International Conference on Pattern Recognition (ICPR) for “How Do Convolutional Neural Networks Learn Design?”