~shahper.shahin
Back to Academic Showcase
Research ExperimentApplied Generative AI • Cipher Schools TrainingJuly 2023

Transformer Architecture Analysis & Prompt Chaining Workflows

Abstract

“Deep dive into Transformer-based LLM architectures, covering multi-head self-attention mechanisms, tokenization semantics, prompt chaining, and function calling workflows.”

PythonLLMsPrompt EngineeringJSON SchemaTransformers

Research / Coursework Objective

Evaluate prompt orchestration techniques (Tree-of-Thought, Chain-of-Thought) for deterministic structured data extraction.

Methodology & Experimental Procedure

  • 1.Theoretical analysis of self-attention matrix computations
  • 2.Comparative evaluation of zero-shot vs few-shot prompt templates
  • 3.Implementation of reusable prompt libraries with JSON schema enforcement

Technical Implementation

Implemented in Python with OpenAI / Gemini API wrappers and Pydantic validation.

Key Findings & Analytical Results

  • Schema-constrained prompt chaining reduces hallucination rates by over 80% on structured outputs
  • Role-prompting with explicit negative constraints yields higher precision on domain tasks
Academic References
  • [1] Attention Is All You Need (Vaswani et al., 2017)
  • [2] Cipher Schools Generative AI Specialization