Samira El-Masri

Systematically decompose complex queries into atomic sub-tasks with dependency mapping and optimized execution planning.

Build a parallel retrieval orchestrator with concurrent source queries, timeout handling, fallback chains, and circuit breakers.

Implement a cross-encoder reranker with batched inference, score calibration, and integration patterns for improved retrieval precision.

Build precise output format specifications combining JSON schemas, natural language instructions, and validation rules for reliable structured outputs.

Build a reasoning trace logger for debugging LLM chains with step tracking, decision logging, and visualization capabilities.

Multi-persona code review approach covering security, performance, maintainability, and reliability perspectives for comprehensive feedback.

Build dynamic retrieval configuration that adapts parameters based on query complexity, expected results, and learned feedback.

Apply chain-of-thought reasoning to system design with structured steps for requirements, components, data flow, and failure analysis.

Design retrieval fusion strategies including RRF, weighted, learned, and cascade approaches for combining multiple retrieval sources.

Few-shot code analysis template with reference examples for systematic evaluation of functionality, complexity, and potential issues.

Build a constrained generation controller enforcing length, format, vocabulary, and structural constraints during and after LLM generation.

Framework for selecting optimal embedding models based on quality, performance, operational metrics, and specific use case constraints.