Engineered a highly performant, custom portfolio website from scratch using the latest Svelte 5 framework. Designed with a zero-clutter and progressive disclosure philosophy, mimicking a technical data sheet.
- Custom Search function using 3-gram Jaccard coefficient.
- Custom CSS-Grid animations for butter-smooth accordion rendering without heavy JS libraries.
- Minimalist 2 layer design: First layer for sharp edges and precision, second layer soft with rounded corners (on hover).
Participated in a competitive university hackathon focused on classifying complex, heavily distorted sensor data using ResNet18 (preTrained then finetuned). Engineered features and trained a robust model under strict time constraints without using guided templates.
- Achieved >88% overall accuracy on the hidden test set, clearly outperforming the baseline.
- Techniques: Ensemlbe, TTA, Image Preprocessing, Data Augmentation
- Secured bonus placement based on model performance metrics.
Pretraining, finetuning and deploying a Tiny YoloV2 on a Nvidia Jetson Nano for Hardware efficient offline Person detection in videostreams.
- Pretraining on ImageNet
- Finetuning using progressive unfreezing and ablation study i.e.comparing different combinations of frozen layers.
- Quantized to int8 and float16 for maximum FPS
- Achieved 42FPS during offline inference
A full-stack MVP web service to extract tables from images and PDFs. Built with a local extraction pipeline capable of running on server CPUs, complete with user and account management. Leverages Mistral OCR for complex tasks.
- Light weight Table extraction from PDF and Image using PaddleOCR
- Runs on CPU - no dedicated server needed
- Mistral OCR for complex etxractions
A locally running tool with a web interface making web research fast and easy. During web research it lets the user save a snapshot of a selected part of a website by pressing a pre configured Keybind. The snapshot is aved into a local DB and can be management in a webinterface. A LLM summarizes the content of the web-page as well as metadata (upload date, url, time of saving). The database lets the user create tags to manage the snapshots.
- Fast and easy web Research
- LLM Summary of source
- Local Database for management of sources
My first contact with RAG and vector databases. It runs a local ChromaDB. The user ingestes pdfs into it. A locally running LLM deployed via Ollama can be asked question based on the pdfs by simple top-k retrieval. It displays the source Page of retrieved information in the Chat interface.
- Retrieval Augmented Generation
- Interactie Chat with source information
- PDF Document ingest
Automated scraping pipeline for Kununu and Trustpilot reviews, generating sentiment analysis using Llama-3.3-70b, BERT, and advanced word matching.
Developed several new Contextual Bandit Algorithms with budget constraints using Neural Networks, Gaussian Processes, Linear Models and Ensembles and Benchmarked in different environments using a facebook CTR dataset.
Fluxio serves as an easily comprehensible and user-friendly interface for visually programming H2T robots, built upon ArmarX. With Fluxio, you can create complex robot capabilities without the need for a deeper understanding of ArmarX or coding skills. Moreover, you should be able to simulate and execute the created programs on real robots.