[iPrakharV]

What changed

• split gateway endpoint logic into small helpers for worker lookup, worker config reads, control calls, cached responses, and prediction calls
• cleaned up the benchmark runner by separating service launch, request sending, latency summaries, and scenario execution
• cleaned up the model evaluation script with typed challenge examples, separate confusion matrix logic, label metrics, and markdown rendering

Why

The project had useful behavior, but a few files were starting to read like long scripts. This keeps the same behavior while making the code easier to scan in a portfolio review.

Validation

• python -m pytest -q
• python -m py_compile on all Python files
• git diff --check
• python bench/run_scenarios.py with output written to /tmp