Mina Ibrahim

I'm an AI Engineer, researcher, and TensorFlow open-source contributor who builds AI systems end-to-end — from model development and experimentation to production-ready applications, APIs, and deployment. I currently contribute upstream to TensorFlow's CPython free-threading enablement work. Multiple changes from this effort have been merged into TensorFlow master, covering CPython 3.14 free-threaded compatibility, native Python extensions, and build infrastructure. As part of this work, I built and validated an experimental TensorFlow build for CPython 3.14t on Linux x86_64 with the GIL remaining disabled across the tested execution path. The work involved auditing native extensions, synchronization, object ownership and lifetimes, callbacks, NumPy C API usage, SavedModel, TensorFlow Lite, checkpoint readers, runtime components, and build tooling. I also identified a free-threading blocker in gRPC's Cython extension, developed and stress-tested an experimental fix under CPython 3.14t, and submitted the change upstream to gRPC. My broader experience spans Machine Learning, Deep Learning, NLP, Arabic NLP, Computer Vision, LLMs, Generative AI, RAG, Agentic AI, synthetic-data generation, model fine-tuning, and production AI deployment. Some of the systems I've built include: • A transformer-based LLM from scratch, including the architecture, attention mechanisms, tokenizer, training pipeline, and real-time inference deployment with FastAPI, along with a 100,000+ sample Q&A training dataset. • Agentic-Nexus, a composable AI architecture combining Multi-Agent orchestration, RAG, CAG/KV-cache-based context reuse, grammar-constrained agent communication, Knowledge Graphs, tool integration, and LLM-as-a-Judge evaluation. • An AI-powered journal management platform that automates submission screening, classification, and reviewer matching using NLP. • Arabic NLP systems using models such as AraBERT and AraGPT2, together with synthetic-data generation using LLMs including LLaMA and BLOOM. • Computer vision and multimodal defect-detection systems using TensorFlow, custom CNNs, Transformers, and OpenCV. • Production-oriented backend and full-stack systems using FastAPI, REST APIs, databases, deployment infrastructure, Linux, and server administration. My current focus is on Agentic AI, LLM systems, ML infrastructure, and open-source AI engineering — including Multi-Agent architectures, RAG pipelines, LangGraph workflows, evaluation systems, inference, APIs, and deployment. My research includes transformer-based NLP, Arabic machine-generated text detection, synthetic conversational datasets, multilingual citation-style classification, dialogue-data processing, and data-driven auditing of university ranking systems. I'm particularly interested in engineering AI systems that connect strong model foundations with reliable, production-oriented infrastructure — from models and reasoning pipelines to native runtimes, APIs, and deployed applications