Yousef Yasser

ˆ Optimized LLM inference to run 20x faster while maintaining output accuracy, improving
production latency and serving cost.
ˆ Designed and built a scalable RAG system retrieving over document layout and visual structure
rather than embedded text, using vector search and embeddings for faster, more accurate, and
eectively unbounded retrieval.
ˆ Integrated LLMs with structured and unstructured document sources to deliver a Trade Finance
product into production for a major bank in Kuwait.

ˆ Deployed state-of-the-art open-source LLMs to production GPU servers with Docker, managing
inference performance and scaling.
ˆ Built RAG applications with reasoning capabilities, implementing embedding and vector-search
retrieval pipelines over open-source models.

ˆ Developed text classi cation models for banking institutions.
ˆ Built LangChain applications leveraging OpenAI GPT models.