GIBU Award 2025 Best Value Award certificate

Avaloq GenAI Corporate Actions for Banking Operations

Agentic AI developed at NEC for Avaloq, decoding SWIFT messages from custodians and populating their corporate actions platform. Avaloq reported 98% accuracy and up to 40% time savings in preparing multilingual notifications. Won the 2025 Datos Impact Award.

Read More

AI4MedCode: AI-Supported ICD Coding for German Hospitals

Developed at QUIBIQ GmbH, this usable NLP solution supports ICD coding and billing in German hospitals. The project was carried out with the University of Stuttgart and Klinikum Stuttgart; my work included German biomedical NER, synthetic clinical text, and Wikidata-based entity linking.

Read More

Kalaam Scribe on an Android phone, with the record control and the transcript visible

Kalaam Scribe

On-device Android dictation in Kotlin and Jetpack Compose. 99 languages via Whisper, plus Roman-Urdu and Urdu-script models. Audio never leaves the phone.

Read More

Earlier work

Recommender System for a German Healthcare Provider

At QUIBIQ GmbH, developed a graph neural network recommender to suggest GOP service codes based on patient ICD diagnosis codes, with patient clustering and interpretable results.

Read More

Smart RFQ for a Global Engineering Services Company

At QUIBIQ GmbH, built an RFQ automation system with SharePoint document ingestion, PDF/Word parsing, TF-IDF matching, and indexed embedding similarity. Deployed on Azure Functions.

Read More

Master’s Thesis: PubMed MeSH Annotation with Graph Neural Networks

Master’s thesis at the University of Stuttgart, carried out at QUIBIQ GmbH. Explored GNN link prediction for assigning MeSH headings to PubMed abstracts, with error analysis of the model’s performance.

Read More

An image passed through the captioning model with a generated English caption

Keras Implementation of Image Captioning Model

Image captioning with InceptionV3 for features and an LSTM for generation, implemented in Keras on a TensorFlow backend.

Read More

Banner for the misinformation spreaders project

Detecting Potential Misinformation Spreaders during Disaster

Finding accounts that spread misinformation during a disaster, so the signal can be separated from the noise while the event is still unfolding.

Read More