feat: refactor devshell as modular standalone path lol

devshells/ml-ai.nix

@@ -0,0 +1,41 @@
+{ pkgs, python ? pkgs.python312, ... }:
+
+let
+  # Create a Python environment with huggingface-cli
+  pythonWithHF = python.withPackages (ps: [ ps.huggingface-hub ]);
+in pkgs.mkShell {
+  name = "ml-ai";
+  packages = with pkgs; [
+    pythonWithHF
+    uv
+    git-lfs
+    ruff
+    pyright
+  ];
+  shellHook = ''
+    # Set up UV
+    export UV_PYTHON="${pythonWithHF}/bin/python3"
+    export UV_LINK_MODE=copy
+
+    # Set up HuggingFace cache (persistent across projects)
+    export HF_HOME="''${XDG_CACHE_HOME:-$HOME/.cache}/huggingface"
+    export HF_HUB_CACHE="$HF_HOME/hub"
+    mkdir -p "$HF_HOME"
+
+    # Optional: Set offline mode (uncomment to use only cached models)
+    # export HF_HUB_OFFLINE=1
+
+    echo "ML/AI shell ready"
+    echo "Python: ${python.version}"
+    echo "HuggingFace cache: $HF_HOME"
+    echo ""
+    echo "Download models with:"
+    echo "  huggingface-cli download <model-id>"
+    echo ""
+    echo "Example TTS:"
+    echo "  huggingface-cli download facebook/mms-tts-eng"
+    echo ""
+    echo "Example LLM:"
+    echo "  huggingface-cli download meta-llama/Llama-3.2-1B"
+  '';
+}