Parallel Gemini API Benchmark (ThreadPoolExecutor)

This project benchmarks and accelerates a 2-step AI pipeline:

1. Image → description (Gemini)
2. Description → 3 quiz questions (Gemini)

Goal: reduce end-to-end latency by parallelizing I/O-bound API calls with Python threads.

What’s inside

• notebooks/parallel_gemini_api_benchmark.ipynb — all experiments, plots, and code
• report/report.pdf — full technical report with results + analysis

Key results

• Serial baseline: ~3.06s per image (100 images)
• Parallel (5 threads): ~4× speedup under free-tier rate limits

Results

How to run

1. Set your API key as an environment variable:
   • export GEMINI_API_KEY="YOUR_KEY"
2. Open the notebook and run all cells.

Notes

• This repo contains no production NeuroTrace code—only the benchmarking pipeline.
• API calls are rate-limited; the notebook includes retry + exponential backoff.