The coding portion included several short exercises for a vision-focused ML role: 1. Given an array of distinct integers, return all possible permutations. 2. Implement a basic grayscale image filter. The input is a 2D integer matrix and a fixed 3 x 3 kernel. Produce the filtered image using convolution, and state clearly how you handle boundary cells. 3. Given a query image embedding and a gallery of image embeddings, return the index of the most similar image. You may use cosine similarity or Euclidean distance, but explain the complexity of your approach and how you would scale it beyond a brute-force scan. Assume standard helper functions for vector math and valid input sizes.