Windows system backup, conflict analysis & containerised AI dev migration toolkit

[Copilot]

Addresses the need to fully snapshot a Windows machine before a clean reinstall, identify dependency conflicts (especially in AI/ML tooling), and boot the new system into a clean, reproducible architecture.

Scripts

• scripts/Analyze-System.ps1 — 17-section system scan: installed apps (winget/choco/scoop/registry), Python/conda envs + package lists, CUDA/cuDNN/GPU, Docker images/containers/volumes, WSL2 distros, VS Code extensions, env vars, PATH, startup programs, drivers, scheduled tasks. Ends with a conflict analysis section that flags multiple Python executables, multiple CUDA installs, conda/pip mixing, and PATH length violations.
• scripts/Backup-System.ps1 — Produces a timestamped restorable bundle: winget/choco/scoop package lists, per-env conda YAMLs, pip freeze, VS Code settings + extensions, SSH keys, Git config, dotfiles, PowerShell profiles, env vars as a ready-to-run restore script, hosts file, scheduled tasks XML, Docker image list (+ optional tar export), optional WSL distro tar exports.
• scripts/Setup-NewSystem.ps1 — Idempotent bootstrap for a fresh Windows install from the bundle. Installs Chocolatey + Scoop, restores packages, enables WSL2 and imports distros, installs Docker Desktop and loads images, recreates conda envs, restores VS Code, SSH, Git, dotfiles, env vars, and scheduled tasks.

Containerised AI Dev Environment (docker/ai-dev/)

Replaces the fragile host-installed CUDA + multiple conda envs pattern with a single GPU-passthrough container. Only the host NVIDIA driver is required on Windows; the CUDA toolkit lives entirely inside the image.

• Dockerfile — nvidia/cuda:12.4.1-cudnn9-devel-ubuntu22.04 base, Miniconda, isolated ai conda env
• docker-compose.yml — GPU passthrough via deploy.resources, JupyterLab (8888), optional TensorBoard (--profile monitoring), optional PostgreSQL (--profile db), named volumes for HuggingFace cache and pip cache
• requirements.txt — PyTorch ~=2.3.0, HuggingFace stack, LangChain, OpenAI, Gradio, Streamlit, FastAPI, W&B, MLflow — pinned with ~= to avoid silent CUDA-incompatible upgrades
• .env.example — Template for HF_TOKEN, OPENAI_API_KEY, etc.; .env is gitignored

Docs

• docs/new-system-architecture.md — Layer diagram (Windows → WSL2 → Docker Engine → per-domain containers), package manager decision matrix, WSL2 .wslconfig tuning, conda env naming conventions, env var hygiene
• docs/ai-dev-environment.md — GPU passthrough prerequisites, CUDA version matrix (11.8–12.4), container workflow, image save/load/push, troubleshooting table

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.