Cairo, Egypt · Reserve Officer (military service completed) · Open to remote & Egypt-based roles
I'm an AI/NLP engineer with ~2 years of experience building speech and language systems, with a focus on Arabic and its dialects. I take models from research notebook to deployed service: fine-tuning Wav2Vec2, Whisper, SpeechT5 and HiFi-GAN, evaluating them properly, then shipping them behind FastAPI + Docker REST APIs.
Latest work
- FluentAI — FastAPI service that scores English pronunciation, fluency and content from uploaded audio, powered by my fine-tuned Wav2Vec2 model.
- MiniRag_26 — step-by-step, production-style RAG backend: FastAPI, PostgreSQL + PgVector, Celery workers, Docker Compose, Prometheus/Grafana, and local LLMs via Ollama.
- BnSR — Bengali.AI Kaggle competition: out-of-distribution speech recognition on 1,200 hours of crowdsourced audio.
What I can build for you
- Arabic / dialect ASR, TTS and full speech-to-speech pipelines
- Automatic video dubbing and speech translation
- Pronunciation and fluency assessment for language-learning products
- RAG and LLM-powered apps, from prototype to a deployed API
- Model optimization for production (ONNX, quantization, parameter-efficient fine-tuning)
- Arabic ASR, TTS, and speech-to-speech pipelines
- Fine-tuning Wav2Vec2, Whisper, SpeechT5, HiFi-GAN (CTC, LoRA/PEFT, layer freezing)
- Pronunciation & fluency assessment (forced alignment, CTC)
- Building production RAG apps and serving models with FastAPI & Docker
- Deep Learning (PyTorch, TensorFlow, Hugging Face Transformers)
- Natural Language Processing (NLP)
- Math for ML (Linear Algebra, Calculus, Probability & Statistics)
- Side projects — Lazy Link (ft. @ElsebaiyMohamed)
- Modablag — End-to-end English→Arabic video dubbing: Wav2Vec2 ASR, neural machine translation, FastSpeech2 TTS, plus punctuation and tashkeel models. Graduation project, graded Excellent with Honors.
- FluentAI — Pronunciation assessment API (Wav2Vec2 + CTC forced alignment, 88% accuracy) returning pronunciation, fluency and content scores.
- MiniRag_26 — Educational production-ready RAG app: FastAPI, PgVector, Celery, Docker, monitoring.
- BnSR — Bengali speech recognition experiments: Wav2Vec2-CTC, transformer decoders, SpeechT5, LoRA/PEFT.
- Speak Lab — Gradio app for English speaking practice with streak tracking and pronunciation scoring hooks, deployed on Hugging Face Spaces.
- Springer, BMC Artificial Intelligence (2025) — co-authored with Dr. Amal.
- A second paper is currently under review in a Q1/Q2 journal.
- Top 18/100 — Arabic Summarization Challenge (ROUGE-L 19.2)
- Bengali.AI Speech Recognition (Kaggle, out-of-distribution ASR) — WER 0.89 · BnSR
|
|
|
