Projects

💬 Given a research paper and a concept, citracer parses the bibliography, identifies the references accross papers (cited near each occurrence) and recursively walks the resulting citation graph. Supports 10+ sources (arXiv, OpenReview, Semantic Scholar, bioRxiv, SSRN, and more), plus optional semantic matching to catch conceptual links the regex misses.

🧠 A flexible machine & deep learning framework built from scratch in pure NumPy, featuring advanced architectures such as Transformers, GANs, variational autoencoders, and bidirectional RNNs. Includes modern optimizers (Adam, RAdam, AdaBelief), advanced loss functions (Focal, Asymmetric), common preprocessing pipelines (BPE tokenization, image augmentation), and much more.

🎵 Self-supervised complex neural network for machine sound anomaly detection using phase-aware spectral analysis. Implements deep complex convolutions with attention mechanisms to capture both real and imaginary components from audio spectrograms, achieving near-SOTA anomaly detection through machine ID classification on industrial sound data.

🌐 A semantic web project enriching Olympics data through spaCy's deep learning models and knowledge graphs. Features enrichments through APIs, SPARQL inference rules, and SHACL constraints. Expanded from 355 to 15,000+ triples with DBpedia/Wikidata linking.

📡 CNN-based bird species audio recognition, optimized using TinyML on STM32 microcontroller with real-time I2S audio processing, 16-bit fixed-point quantization, and LoRaWAN IoT connectivity achieving 86% accuracy with optimized memory footprint for embedded deployment.

🦠 Python script that simulates a pandemic using moving dots with tweakable parameters and then generates a graph according to obtained values. Validated against research literature with real-time intervention scenario analysis.