Arij Soula

– Developed an NLP-based code similarity detection system using transformer models to identify semantic patterns in large-scale codebases. – Built an end-to-end ML pipeline from data preprocessing to model deployment, processing 100K+ code samples with vector embeddings. – Implemented a scalable vector database solution using Qdrant for fast and efficient code similarity matching across enterprise-scale repositories. – Technologies: Python, PyTorch, Hugging Face, Qdrant, Data Retrieval Systems

• Designed and deployed end‑to‑end speech recognition pipelines leveraging transformer‑based architectures (BERT, Whisper, Wav2Vec2) for accurate call transcription and voice analytics • Fine‑tuned STT and TTS systems, improving transcription accuracy by 15% and speech naturalness. • Tech Stack: Python, PyTorch, Transformers, Whisper, BERT, Azure ML, NLTK, Speech AI, NLP