Fixed-scope review of a Python/numpy codebase for a scientific experiment

I'm looking for an experienced Python developer to review about 300 lines of code from a scientific experiment with a large language model (LLM). The code picks each word of the generated text using a random number. Three sources for that number are compared, and the code must treat all three identically, changing only the source. The review covers three points, described in detail in a scope document: Word selection: cumulative sum of probabilities in float64, with canonical ordering, slice boundaries, and a second stage for the tail. Order of operations: making sure that bits read from a USB device are generated after each word's options exist, and that no old data is reused. The pseudorandom control group: confirming that it uses no system entropy during sessions. The code is commented (in Portuguese), has 31 automated tests, and includes a README. Profile: Python with numpy, with attention to numerical detail (floating point, ordering, reproducibility). Nice to have: PyTorch or Hugging Face Transformers, and experience with scientific code. Deliverables: Line-by-line comments on a GitHub pull request (private repository). A short list of findings, each marked critical, important or minor. How we work: the code was written with an AI research assistant. Technical questions are answered in writing by the assistant, through the project coordinator. Estimate: 4 to 6 hours, including reading the context, running or writing a few tests of your own, and writing up the findings. Fixed price. Confidentiality: the code and the scope document are shared after hiring.