Skip to content

Intelligence Profiles Demo Script ​

Six demos. About fourteen minutes. Company-wide context and term registries combine with each user’s role and personal terms. GhostAI injects the company Intelligence Profile into LLM system and tool prompts (cached about five minutes), merges org and user terms for detection, and records auditable settings changes for key org operations.


Pre-Demo Checklist ​

  • [ ] GhostAI backend and frontend are running and reachable (for example backend on :8000)
  • [ ] Demo account belongs to an organization (tenant context required for merged terms and org profile APIs)
  • [ ] For company profile and org terms: demo user has Owner or Admin (read for GET; write for PUT, POST auto-research, terms, bulk)
  • [ ] Browser signed in; Settings (or Admin) includes Company and Personal Intelligence Profile tabs (CompanyProfileTab, PersonalProfileTab)
  • [ ] Optional: API key in GHOSTAI_API_KEY if you will show curl for GET /api/intelligence-profiles/merged-terms
  • [ ] Optional: SIEM or database access if you will mention audit events (settings.update with details.section values below)
  • [ ] Know your demo organization ID (Admin → Organization or API) for narrating paths like /api/intelligence-profiles/company/{org_id}

The Story ​

Riley is an IT admin rolling out GhostAI. The org needs consistent handling of industry-specific jargon and internal codenames, without blocking public product names. Jordan is a compliance analyst who tightens personal sensitivity and adds a few role-specific terms. Together, company and personal profiles shape detection and what the model “knows” about the business.


Demo 1: Company Profile Setup ​

Duration: 3 minutes
Goal: Show structured org context and the sensitivity floor in the UI.

Steps (Company Profile tab) ​

  1. Open Company Intelligence Profile (or org settings area that hosts CompanyProfileTab).

  2. Set or edit:

    • Company description — maps to company_description (free text).
    • Industry details — structured fields stored as industry_details (object/dict), for example sector, regulated flag, frameworks.
    • Products and services — products_services (list; entries may be strings or objects depending on what you paste or import).
    • Data classification tiers — data_classification_tiers (list describing how the org labels data).
    • Sensitivity floor — sensitivity_floor: one of relaxed, balanced, paranoid (org-wide minimum posture for detection context).
  3. Save and point out profile completeness if shown (backend computes profile_completeness on save).

Talking Points ​

"This is not just marketing copy. We serialize the company profile into context the product can use, and we cache injected company context for about 300 seconds so chat stays fast while staying fresh."

"The sensitivity floor is the organization’s baseline. Personal profiles can dial the user up, but not below what the org mandates."


Demo 2: Auto-Research ​

Duration: 2 minutes
Goal: Show AI-assisted population from company name and domain without auto-saving.

Steps ​

  1. In the company profile flow, run Auto-research (calls POST /api/intelligence-profiles/company/{org_id}/auto-research).

  2. Enter a real company name and domain (for example Northwind Traders and northwind.example).

  3. Show the response shape: suggested company_description, industry_details, products_services, data_classification_tiers, suggested_sensitivity_floor, suggested_sensitive_terms, suggested_safe_terms, sources, confidence, warnings.

  4. Apply chosen fields manually via the normal PUT company profile and POST term endpoints (the research endpoint does not persist).

Talking Points ​

"We pull public signals—search, site content, synthesis—and return suggestions with confidence. Your admin decides what lands in production. That keeps governance in human hands."

"If you omit the name, we can fall back to the org name or infer domain from admin email—handy for quick trials."


Demo 3: Custom Terms ​

Duration: 2 minutes
Goal: Sensitive vs safe terms, metadata, and actions.

Steps ​

  1. Add an org sensitive term (for example an internal codename). Fields map to API TermCreate:

    • term — phrase to match (word-boundary sensitive matching in detection).
    • term_type: sensitive
    • category — e.g. project, general
    • sensitivity_action: always_protect, context_dependent, or never_protect
    • context_notes — optional analyst note
  2. Add an org safe term (for example a public product name customers say daily):

    • term_type: safe
    • Same category, sensitivity_action, and context_notes as needed.
  3. Explain behavior:

    • Sensitive terms produce CUSTOM_SENSITIVE entities in the orchestrator, word-boundary match, score 0.95, detector_source: intelligence_profile.
    • Safe terms suppress detections whose entity text is an exact, case-insensitive match to the safe string (after other detectors run).

Talking Points ​

"Safe terms are surgical: the span has to match exactly. That avoids accidentally unmasking adjacent PII in the same phrase."


Demo 4: Personal Profile ​

Duration: 2 minutes
Goal: User-scoped context in PersonalProfileTab.

Steps ​

  1. Open Personal Intelligence Profile.

  2. Set:

    • job_title
    • department
    • work_description
    • data_types_handled (list, e.g. ["PHI", "customer PII"])
    • sensitivity_level: relaxed, balanced, or paranoid
  3. Note onboarding state (onboarding_dismissed) if you demo the wizard.

Talking Points ​

"Personal profile captures how this human works with data. We merge their term list with the org’s for detection, with a clear rule: you cannot mark a term safe personally if the org already registered it as sensitive—the API returns 409 Conflict with an explicit message."


Demo 5: Live PII Detection Impact ​

Duration: 3 minutes
Goal: Prove custom terms affect preview detection the same way as chat.

Steps ​

  1. Ensure Demo 3’s sensitive term is saved (example codename: NighthawkProtocol).

  2. Open PII preview or use the authenticated preview endpoint POST /api/detect with body {"text": "…"}.

  3. Send text that includes the codename as a standalone token, e.g. Status: NighthawkProtocol is approved for pilot.

  4. In the entity list, show:

    • entity_type: CUSTOM_SENSITIVE
    • detector / detector_source: intelligence_profile
    • score / confidence: 0.95
  5. Send text that includes only a safe-listed term as its own detected span (e.g. a brand name you put on the safe list that would otherwise be flagged as an organization). Show that entity disappears after safe-term suppression.

Talking Points ​

"This is the same orchestrator path we use in the product: merge org + user sensitive lists, match with word boundaries, then strip anything that exactly matches a safe term."


Demo 6: Merged Terms Preview ​

Duration: 1 minute
Goal: One read model for effective lists.

Steps ​

  1. Call GET /api/intelligence-profiles/merged-terms with the user’s session or bearer token.
  2. Show JSON:
    • sensitive: string array (org + personal combined, deduplicated)
    • safe: string array (org + personal combined)

Example (narrate, do not read every token) ​

bash
curl -s -H "Authorization: Bearer $GHOSTAI_API_KEY" \
  https://api.example.com/api/intelligence-profiles/merged-terms | python3 -m json.tool

Talking Points ​

"Downstream detection does not need to understand two databases—this endpoint is the merged effective policy for this user in this org."


Audit & Compliance (Elevator) ​

When you need a one-liner for security buyers:

  • Org profile PUT, auto-research, single term POST, bulk POST .../terms/bulk, and term DELETE emit audit via settings.update on resource type organization, with details.section one of:
    • company_profile
    • auto_research (includes researched company_name / domain in details)
    • org_terms
    • org_terms_bulk (includes inserted and total)

Personal profile PUT and personal term POST in the current API do not add the same structured audit rows as those org endpoints; lead with org-side auditability for SOC2-style narratives.


Timing Cheat Sheet ​

DemoMinutes
1 Company profile3
2 Auto-research2
3 Custom terms2
4 Personal profile2
5 Live detection3
6 Merged terms1
Total~13

Add buffer for questions: 12–16 minutes total.

SOC 2 Type II Certified | BAA Available | HIPAA-Ready