Peer Review

One-time computer setup

Complete these steps once on each Windows computer that will analyze drawings.

1. Open the Ollama for Windows download, select Download for Windows, and run OllamaSetup.exe. Use the normal installer; the standalone ZIP is not needed.
2. Install the LTS version from the Node.js download page. Keep the normal installation choices.
3. Open d3adgoose/NSAutomation on GitHub. Select the green Code button, then Download ZIP.
4. Open the downloaded ZIP file, select Extract all, and save the extracted folder somewhere permanent, such as Documents. Local AI needs this folder to stay in that location.
5. Open the extracted NSAutomation folder and double-click Set Up NS Local AI.cmd. If Windows asks for confirmation, confirm only after checking that the file came from the repository linked above. Setup will try to approve the N/S website automatically; on managed computers it safely continues with manual Chrome permission instead.
6. Leave the setup window open while it downloads qwen3-vl:8b-instruct for fast checks and qwen3-vl:30b-a3b-instruct for difficult visual checks. The quality model is about 20 GB, so the first setup may take a while. If that download cannot finish, Peer Review continues with the fast model and setup can resume the quality download later.
7. Close and reopen Chrome, open this website, and select Log in. Sign in with your normal company account.
8. Open the Local AI popup and select Try Reconnecting. If a Chrome permission question still appears, select Allow. Drawing files remain local.

After setup: select the Local AI status button. A green circle means the service and fast model are ready. The popup lists the quality model when it is installed. You do not need to run the setup file again each day.

Model files and Git: Ollama normally stores models in C:\Users\your-name\.ollama\models, outside the NSAutomation folder. Do not copy that folder into the project or upload it to GitHub. The repository's .gitignore also excludes local model folders, generated review files, logs, and AutoCAD recovery files.

If setup cannot find Ollama: finish running OllamaSetup.exe, then close and reopen Set Up NS Local AI.cmd. If the model download stops, run setup again; completed model data is reused.

Local AI popup: follow the displayed Your next step message. Use Try Reconnecting after correcting an issue, or Close to return to Peer Review.

Using Local AI for Peer Review

Upload the clean original engineering PDF and select Run Automatic Checks or Validate. OCR reads image-only title blocks. The fast local Qwen model handles routine passes, while the quality model handles dense transcription, specialist sweeps, and source verification. A composite DWG sheet can activate drawing, equipment, plumbing, and electrical sweeps together. Native fitting/component and electrical-run audits are counted separately; quantity or callout differences remain engineer-review prompts unless exact drawing evidence proves a conflict. Activity messages show which model completed each difficult pass. Existing reviewer annotations are not required. Keep the page open until the activity message says the automatic review is complete.

Every result remains pending until you review it. Confidence shows the model's certainty, while Rule Check identifies a direct comparison performed without AI judgment.