Samira El-Masri

Few-shot query classification with examples for factual, procedural, analytical, comparative, and exploratory categories with retrieval recommendations.

Optimize model serving costs through infrastructure, model, traffic, and architecture optimizations while maintaining SLA compliance.

Zero-shot entity extraction using only type definitions with span identification, confidence scoring, and disambiguation handling.

Build a prompt template library with versioning, variable validation, template inheritance, and usage analytics.

Apply tree-of-thought reasoning to systematically compare ML architecture options with pros/cons analysis and decision matrices.

Design an embedding update pipeline with stale detection, priority-based processing, batch updates, and quality validation.

Apply reflection-based error recovery with systematic failure analysis, diagnosis, strategy selection, and post-recovery learning.

Build a context relevance scorer combining similarity, keyword, entity, and topic signals to filter retrieved documents before LLM generation.

Apply chain-of-thought debugging with systematic symptom analysis, hypothesis generation, evidence collection, and root cause identification.

Design comprehensive health checks for AI infrastructure covering embedding services, vector DBs, and LLM APIs with aggregated status.

Build a self-consistency voting aggregator with majority voting, semantic clustering, and confidence-weighted consensus extraction.

Tune vector similarity thresholds through systematic evaluation of precision, recall, and F1 scores with distribution analysis.