AI Engineer Assessment Pipeline

Overview

An email generation assistant that benchmarks two LLMs head-to-head — Llama 3.3 70B vs Gemma2 9B via Groq — using a custom evaluation framework with cross-model judging to eliminate self-grading bias.

Key Features

• Streamlit UI for email generation and side-by-side model comparison
• Custom eval metrics: Fact Recall, Tone Alignment, and hybrid Fluency/ROUGE-L
• LLM-as-a-Judge pattern with cross-model judging to avoid self-grading bias
• Groq inference for fast evaluation cycles across both models
• Structured evaluation report with per-metric scores and reasoning