Reem Hassoun

Designed and implemented multiple production generative-AI applications: text2sql, legal Q&A, recommendation engines, plagiarism detection, and grammar/spell checkers for Arabic and English. Built an agent-based system including hierarchical agents and integrated it with the A2A protocol within mulit-projects. Built hybrid retriever + reranker pipeline using Haystack and Neo4j to support accurate, high-throughput RAG for legal projects; improved answer relevance by 15%. Implemented model optimization strategies: QLoRA finetuning, bitsandbytes quantization, and DeepSpeed for multi-GPU distributed training to reduce inference latency and GPU memory usage; achieved measurable speed and memory improvements. Integrated LLMs with NLU to build hierarchical agent-based systems for self-investigation and virtual legal assistant products. Developed OCR + vision pipelines (e.g., Qwen-VL) for forensic video analysis and automated incident reporting extraction. Built video and meeting summarization systems leveraging STT + LLM summarizers and deployed services in Dockerized microservices. Wrote production-ready Python modules and APIs (Flask) for exported agent logic as reusable Python code for engineering teams.

Taught Python, C++, Java, Database (Oracle), Distributed Mathematics, and Algorithms & Data Structures to undergraduate students. Designed curricula and assignments emphasizing software engineering best practices and algorithmic thinking.

Delivered courses on advanced programming (C++, Java), MATLAB, data structures & algorithms, and theoretical Computer Science topics to university students and professionals.