Samira El-Masri

Implement reliable structured output enforcement using prompt engineering, constrained decoding, validation, and LLM-based repair strategies.

Systematically optimize vector search latency through profiling, index tuning, query optimization, and caching with quantified trade-off analysis.

Build a dynamic few-shot example selector using semantic similarity, diversity optimization, and quality scoring for improved LLM task performance.

Design comprehensive prompt caching strategies analyzing prefix, response, and KV cache opportunities with provider-specific optimizations.

Conduct a comprehensive RAG system audit covering retrieval quality, generation fidelity, operational health, and security compliance.

Structured incident response guidance for AI system production issues with immediate actions, investigation steps, and communication templates.

Build a smart model router that dynamically selects optimal LLMs based on query complexity, cost constraints, and capability requirements.

Implement context compression techniques including summarization, query-focused extraction, and token pruning to maximize information density.

Set up a comprehensive LLM observability stack covering logging, metrics, tracing, and alerting with tool recommendations and configurations.

Plan embedding model migration with dual-write strategy, validation checkpoints, traffic shifting, and rollback procedures for zero-downtime updates.

Build a synthetic data generator for RAG testing covering document generation, QA pairs, and edge cases with quality validation.

Design multi-stakeholder retrieval metrics dashboards for engineering, quality, and business audiences with specific metrics and alert configurations.