Train the AI (Expert Center)
Two independent layers — what your team knows, and how your team thinks. This is the highest-leverage screen in DarkBird.
Found in Settings → Expert Center. Available to Owners and Admins. This is where you give the AI the knowledge it needs to estimate, scope, and ask the right questions for your specific business. It works through two independent layers — Knowledge Categories and Expert Profiles.
Knowledge categories and expert profiles
| Layer | What it contains | When the AI reads it |
|---|---|---|
| Knowledge Categories | Product facts, labor benchmarks, install notes, field conditions, manufacturer quirks — the "what to know" for each trade area | Fresh on every interview turn and every card suggestion |
| Per-category Expert Profiles | How your best engineers think about a specific trade area — what they probe, how they estimate, what makes them nervous | Before every interview turn in that category — shapes how the AI probes, not just what it knows |
| Org-wide Expert Profile | Cross-cutting instincts that apply across all trade areas — phasing philosophy, subcontracting rules, client communication style, non-negotiables | Every interview turn and card recommendation, regardless of trade area |
They are independent. Add a new spec sheet to a Knowledge Category — the Expert Profiles do not change. Update an Expert Profile when your team's thinking evolves — the Knowledge Categories stay intact. Together they make every AI interaction reflect how your team actually works.
Adding content to a category
- Add Text (inline)
Click Add Text directly in the source panel to type or paste notes — labor rates, house standards, lessons learned, or anything from your clipboard. No dialog needed. This is the fastest path for quick additions.
- Upload a File or Scrape a URL
Click Add Source then choose Upload or Scrape URL. Supports Excel, CSV, PDF, Word, TXT, and Markdown files. Paste any public URL — a distributor spec page, manufacturer guide, or NSCA resource. The AI fetches the page and extracts install-relevant content.
- Research
Click Research in the category action bar to pull from web sources automatically. The AI searches for current install practices, labor benchmarks, and field notes relevant to that trade area and merges new findings into your knowledge. Use this to refresh a category that has not been updated in a while.
- Product Lookup
Click Product Lookup and enter a manufacturer, model, and product name. The AI runs deep research on that specific product — installation requirements, integration gotchas, wiring specs — and adds the findings to that category's knowledge. Use this when you are about to quote a product your team has not installed before.
After adding, the source appears in the list with its name, type, character count, and date added.
Removing a source
Hover over any source card and click the trash icon to remove it. A Sync prompt will appear — click it when you are ready to update what the AI reads.
Syncing knowledge
After adding or removing sources, an amber "Sources changed — click Sync" banner appears. Click Sync to synthesise all remaining sources into a fresh knowledge snapshot. This is the version the AI reads during interviews — not the raw sources themselves. A green "AI knowledge is up to date" indicator confirms when the AI is current.
Start with your labor rate sheet and one past project quote. Even a small, accurate Expert Center significantly improves first-estimate accuracy. Use Research or Product Lookup to fill gaps as new product lines arrive.
Expert profiles
Expert Profiles capture how your best engineers think — not what they know. There are two scopes, and they work together:
| Scope | What to put here | Where to find it |
|---|---|---|
| Per-category | Trade-specific instincts — what a camera engineer always asks, how a network engineer estimates switch ports, what an access control tech checks before committing to a head-end size | Select a category in the left panel, then click the Expert tab |
| Org-wide (cross-cutting) | Instincts that apply across all trade areas — phasing philosophy, subcontracting rules, site-survey minimums, coordination buffers, client communication style | Select Expert Profile at the top of the left panel |
When the AI conducts an interview, it reads the per-category expert for the active trade area and the org-wide profile together. Build the per-category profiles to sharpen domain-specific probing; build the org-wide profile to shape the overall approach.
Interview your expert
Click Interview Expert to open an 8-question wizard. The org-wide profile and per-category profiles ask different sets of questions, because they shape different things.
Org-wide Expert Profile covers how your firm operates across every project type: discovery process, firm-wide standards, overhead & margin, scope limits, change orders, team structure, client dynamics, and competitive edge.
Per-category Expert Profiles cover domain-specific technical reasoning: mental model, rules of thumb, non-negotiable questions, handoffs, where scope bleeds, preferred vendors, house rules, and hard limits.
Skip any question that does not apply — the AI builds the profile from whatever you provide. The more specific your answers (real numbers, real vendor names, real site conditions), the more accurate the AI's estimates. After the wizard generates a draft, you can edit it before applying. Profiles are stored in plain text and can be refined at any time.
When to update
- After winning or losing a major bid where your scope strategy was the deciding factor
- After onboarding a new specialty (e.g. you just hired your first DAS engineer — build that category's expert profile)
- When your preferred vendors change or a manufacturer's commissioning process shifts
- After a project that taught your team something the AI does not know yet
Interview feedback — refine from real sessions
Every time an Owner or Admin uses the thumbs-down button during an interview to flag an irrelevant question, that signal is stored here under Interview Feedback. The panel shows the number of unprocessed flags waiting to be acted on.
Click Refine from Feedback to run the AI over your accumulated flags. It reads each dismissed question alongside the conversation context at the time — understanding why the question was wrong for that project type — and updates your org-wide Expert Profile's reasoning to avoid the same error in future sessions.
How it works: This does not add a blocklist of banned questions. Instead, it refines the reasoning pattern — for example, teaching the profile to distinguish between "Wi-Fi mentioned as a VLAN segment in a firewall project" vs. "Wi-Fi as a deployment in scope." The result is a smarter expert, not a longer prompt.
Custom knowledge categories
In addition to DarkBird's built-in categories (Camera/Surveillance, Network Infrastructure, etc.), you can add your own to cover specialties not in the standard list — for example "SaaS Management", "Subcontractor Rates", or "Client Preferences". Click Add Expert at the bottom of the left sidebar to create one.
Deleting a custom category
Hover over the category name in the sidebar and click the ×, or select the category and click Delete in the action bar. Deleting removes the category and all of its knowledge documents, including its Expert Profile. This cannot be undone.
Standard categories cannot be deleted. Delete and Merge controls only appear for categories you created under the Custom section.
Merging a custom category into another
If a custom category duplicates or overlaps an existing one, you can merge it rather than delete it. Click Merge into… in the action bar, choose the target category, and confirm. All knowledge documents move into the target and the custom category is deleted.
Note on Expert Profiles: The category's Expert Profile text is not automatically merged — the AI's reasoning persona requires human judgment to combine cleanly. If the custom category has a meaningful Expert Profile, open the target category's Expert tab and copy the relevant text in manually.