Get the best out of NetArchitect AI

1. From Workspace, create a new project with a customer name and short description.
2. Open the project and complete the Intake form — the AI uses every field you fill in.
3. Click an artifact in the left rail (HLD, LLD, BOM, …) to generate it.
4. Pick the right industry template from the dropdown above the artifact.
5. Use Preview to validate, then export to .docx, .pdf or .xlsx.

The intake is the prompt. Every paragraph you skip becomes a hallucination later. Aim for:

• Business drivers — name the outcomes (e.g. “sub-50ms RTT to AWS for trading apps”).
• Success criteria — measurable, not aspirational.
• Scale — sites, users, racks, throughput in Gbps, growth horizon.
• Constraints — vendor preferences, existing standards, budget bands, change windows.
• Compliance — PCI, HIPAA, NIS2, FedRAMP — mention every framework that applies.

Rule of thumb: if a senior architect would ask “how many?” or “what version?”, the AI will too — answer it up front.

• HLD — for executives & solution review boards. Logical view, intent, risks.
• LLD — for implementation engineers. VLANs, routing, configs, validation.
• BOM / BOQ — for procurement. SKUs, optics, licensing, quantities, prices.
• BOD — for the design authority. Standards, decisions, assumptions log.
• MOP — for change advisory. Pre-checks, cutover steps, rollback.
• Test plan / Runbook — for operations. Acceptance criteria & day-2 procedures.

The industry preset shapes the cover page, footer classification and framework citations. Pick the closest vertical or build a custom preset:

• Use Auto-detect to let the AI rank presets against the artifact content — review alternates before applying.
• Use Manage presets to create custom verticals with your own banner, frameworks and per-artifact subtitles + section labels.
• Use Copy mappings from… to merge or replace per-artifact mappings from another preset, with a side-by-side diff before confirming.

The Preview dialog renders the document the same way the export will. Check:

• Cover banner & classification footer are populated.
• The TOC numbers match the section outline.
• BOM/BOQ tables render with proper headers in .xlsx.
• Validation warnings appear at the top of the dialog when fields are missing.

• tip
  “Assume Cisco Validated Designs unless otherwise noted.”
• tip
  “Use VXLAN/EVPN with anycast gateway; avoid legacy STP.”
• tip
  “Quantities and SKUs must reconcile to BOQ totals.”
• tip
  “Cite each section to a specific framework (PCI 1.1.x, NIST 800-53 SC-7…).”
• tip
  “Provide rollback steps before validation steps in any MOP.”
• tip
  “Group BOM by site, then by role (spine/leaf/border).”
• tip
  “Show oversubscription ratio per pod.”
• tip
  “Call out single points of failure in a dedicated risks table.”
• tip
  “Express targets as SLOs (e.g. P99 < 2 ms intra-DC).”
• tip
  “Prefer tables over prose for any list of more than 3 items.”

• Banner empty — open Manage presets and set a cover banner.
• Classification missing — required for the page footer in every export.
• No frameworks listed — warning only; cover page will look sparse.
• Excel sheet name >28 chars — shorten the preset short name.
• Banner >60 chars (DOCX) — Word will wrap awkwardly; shorten it.

Auto-detect sends the artifact title, type and content to the AI which ranks every available preset (built-in + custom) by fit, returning a confidence score and one-sentence reason for each.

Use the confidence bar to compare alternates and click Apply on any row to switch templates. Detection never overwrites your selection without confirmation.

1. Open Edit intake and add specifics: site count, throughput, vendor preference.
2. Re-generate the artifact — it picks up the new intake immediately.
3. If the answer still feels thin, switch to a more specific industry preset.

Custom presets live in your browser and travel via JSON export/import. To build one:

1. Open Manage presets → New preset.
2. Set label, banner and classification (required).
3. Use Copy mappings from… to seed per-artifact subtitles & sections from a built-in.
4. Tweak per-artifact rows and Save.
5. Click Export JSON to share with teammates.

• Always finish intake before generating — never half-fill it.
• Generate HLD first; use it as the source of truth for LLD/BOM.
• Run Auto-detect on every artifact to confirm the template fit.
• Preview every export before sending externally.
• Save customer-specific presets for repeat engagements.
• For sensitive customers, set the classification line to match their data-handling agreement.

• Lead with business intent + non-functional targets (RTO, RPO, SLO).
• Use logical zones (DC fabric, edge, WAN, branch) before product names.
• Always include a risks & assumptions table.

• Reference the parent HLD section for every block.
• Include validation commands per device class.
• Avoid prose for tables that fit better as structured data.

• Group rows by site, then by role (spine/leaf/border/firewall).
• List optics & licenses on separate lines, not bundled in the chassis row.
• Quantities must reconcile across HLD/LLD/BOM.

• Define rollback before validation; CAB requires it.
• Time-box each step; no open-ended “verify” bullets.
• Include a backout decision tree if cutover fails midway.