Samira El-Masri

Zero-shot classification prompt with systematic feature extraction, category matching, and confidence assessment for accurate categorization.

Build a comprehensive chunking quality evaluation system measuring coherence, completeness, boundary quality, and retrieval effectiveness.

Implement contextual bandits for adaptive LLM model selection using LinUCB or Thompson Sampling with online learning updates.

Design a token budget allocation system for multi-step LLM pipelines with dynamic rebalancing and priority-based distribution.

Create meta-prompts that dynamically adapt to new tasks through self-analysis, strategy selection, and verification steps.

Build a comprehensive LLM cost attribution system with multi-dimensional tracking, budget alerts, and internal billing capabilities.

Integrate sparse embeddings with dense embeddings for hybrid search using BM25/SPLADE with configurable score fusion.

Apply reflection-based prompt optimization through iterative analysis, hypothesis generation, revision, and meta-reflection cycles.

Build a query intent classifier to optimize RAG retrieval strategies based on factual, comparison, summarization, and other intent categories.

Generate customized production rollout checklists for AI features covering pre-deployment, deployment, validation, and post-deployment phases.

Build latent space interpolation tools for exploring semantic gradients, concept blending, and corpus gap analysis in embedding spaces.

Apply tree-of-thought reasoning to decompose complex problems with systematic branching, evaluation, and convergence on optimal solutions.