feat: Add multi-model LLM benchmarking notebook

[xeroxpro]

This PR introduces a Jupyter Notebook (.ipynb) that provides an end-to-end framework for evaluating and benchmarking multiple Large Language Models from various providers.

What this notebook contains:
Dynamic Client Initialization: The notebook automatically detects available API keys (OpenAI, Groq, DeepSeek) from environment variables and checks for local services (Ollama) to dynamically configure the list of competitor models.

Weighted Question Generation: Implements a two-step process to generate the evaluation questions:

Generates three distinct, nuanced questions.

Uses a "Consultant" LLM (gpt-4o-mini) to rank these questions and assign weights (50, 30, 20 points).

Parallel Query Execution: A loop iterates through all configured models and collects their responses to all three weighted questions.

Dual-Weight "Judge" Scoring:

All responses are formatted and sent to a single "Judge" LLM (gpt-4o).

The Judge provides two scores for each answer: a judge_score (60% weight) and a peer_average_score (40% weight).

Final Score Calculation: A script parses the Judge's nested JSON output, applies the answer-level (60/40) weights, and then applies the question-level (50/30/20) weights to calculate a final score for each model.

Results Visualization: Uses matplotlib to generate a horizontal bar chart of the final results, highlighting the winning model in a different color.

Dependencies:
The notebook requires the following Python packages:

openai

requests (for the Ollama health check)

matplotlib (for the final graph)

numpy

jupyter

[Copilot]

Copilot wasn't able to review any files in this pull request.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ed-donner]

Hey - this is great - but would you be OK to clear the outputs first for me? Otherwise it's 2,000+ loc