Samira El-Masri

Build a smart context window manager handling token budgets, content prioritization, and overflow strategies for optimal LLM context utilization.

Build a comprehensive RAG evaluation framework measuring retrieval, generation, and end-to-end quality with automated and human evaluation protocols.

Design a multi-model fallback orchestration system with circuit breakers, prompt adaptation, and quality monitoring for high-availability LLM applications.

Build a production-ready LLM streaming response handler with backpressure management, error recovery, and real-time metrics collection.

Optimize vector index configurations by analyzing HNSW and IVF parameters with performance trade-off analysis and memory impact assessment.

Systematically analyze end-to-end latency in ML pipelines to identify bottlenecks and prioritize optimization efforts by impact and implementation effort.

Build a hybrid search system combining dense embeddings and sparse retrieval with score fusion and reranking for improved retrieval accuracy.

Optimize LLM prompts for token efficiency across multiple risk levels with specific reduction strategies and trade-off documentation.

Build a production-ready LLM request batching system with dynamic sizing, priority queues, and comprehensive error handling for cost and throughput optimization.

Optimize document chunking strategies for RAG systems through systematic analysis of chunk sizes, overlap, and semantic boundaries with implementation code.

Systematically analyze RAG retrieval quality through structured evaluation of results, failure patterns, and improvement hypotheses with A/B test proposals.

Design comprehensive ML system observability dashboards covering real-time operations, performance analysis, and cost monitoring with specific visualization configurations.