Appearance
GhostAI in 18 Minutes — Master Demo
This is the flagship demo for GhostAI. In 18 minutes, walk through the entire platform: real-time PII detection, secrets scanning, privacy-respecting memory, multi-model governance, enterprise compliance controls, industry vertical switching, and geographic intelligence with optional Google Maps Premium. Ideal for first meetings with CISOs, CTOs, and compliance leaders across any vertical.
Pre-Demo Checklist
- [ ] GhostAI instance running with demo data
- [ ] No vertical pre-selected (default mode)
- [ ] Healthcare vertical available for Chapter 6
- [ ] Audit log and SIEM panels accessible in admin
- [ ] Comparison mode enabled
- [ ]
GEO_PROVIDER=hybridandGEO_GOOGLE_MAPS_API_KEYset in.envfor Chapter 7 - [ ] User account on Enterprise tier (or
geo_google_mapsfeature flag enabled for demo tier)
Demo Script
Chapter 1: PII Detection
Duration: 2 minutes Story: A routine customer follow-up email exposes SSN, credit card, address, and phone number. GhostAI detects and tokenizes all PII before the request reaches the LLM.
Input:
Please help me draft a customer follow-up email. The customer is John Martinez, SSN 425-68-9103, lives at 742 Evergreen Terrace, Springfield IL 62704. His account number is 4532-8891-2201-3847 and his phone is (312) 555-0189. He called about a billing discrepancy of $2,847.50 on his October statement.Expected Output:
- PII detection badges appear: SSN, credit card, address, phone, name
- Tokenized version shown in the "What the AI sees" panel
- LLM responds with a professional follow-up email using tokenized placeholders
- Detokenized response returns real values to the user
Talking Points:
"Every piece of PII — the SSN, credit card, address, phone number — was detected, tokenized, and stripped before it ever reached the AI model. The model composed a professional email without seeing a single real data point. That's GhostAI."
Chapter 2: Secrets Detection
Duration: 2 minutes Story: A developer pastes a Python configuration file containing AWS keys, a Stripe secret key, and a database connection string. GhostAI identifies and blocks all credential leakage.
Input:
Review this configuration for security issues:
```python
import boto3
AWS_ACCESS_KEY_ID = "AKIAIOSFODNN7EXAMPLE"
AWS_SECRET_ACCESS_KEY = "wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY"
STRIPE_SECRET_KEY = "sk_live_4eC39HqLyjWDarjtT1zdp7dc"
DATABASE_URL = "postgresql://admin:SuperSecret123@prod-db.internal:5432/main"
client = boto3.client("s3")
```Expected Output:
- Secrets detection fires: AWS access key, AWS secret key, Stripe key, database URL with credentials
- Each secret is tagged with its type and severity
- LLM receives sanitized code with secrets replaced by safe placeholders
Talking Points:
"This isn't just PII — these are live credentials. An AWS key pair, a Stripe secret, a database password. GhostAI caught them all before the model saw a single character. For engineering teams, this is the difference between a security review and a breach."
Chapter 3: Memory That Respects Privacy
Duration: 3 minutes Story: Alex Rivera introduces himself as the new CISO at Northwind Financial, sets communication preferences, then starts a new conversation. GhostAI remembers the preferences — but stores only the sanitized version.
Input (first message):
I'm Alex Rivera, the new CISO at Northwind Financial. We're standardizing on EU data residency for all AI workloads and I prefer concise bullet-point answers with executive summaries. Remember this for future conversations.Input (new conversation):
Draft an internal memo about our AI security policy. Keep it aligned with my preferences.Expected Output:
- First message: GhostAI confirms memory stored with PII-safe content
- Second message (new conversation): Response uses bullet points and executive summary style
- Memory Manager shows the stored preference with sanitized content
Talking Points:
"Alex told GhostAI his name, role, company, and preferences. In the next conversation, the AI remembered his formatting preferences — bullet points, executive summaries — without storing raw PII. The memory is useful, but the vault is clean."
Chapter 4: Multi-Model Governance
Duration: 2 minutes Story: The same prompt is sent to multiple AI models simultaneously via comparison mode. All models receive the same tokenized input — no model gets raw data, and every response is audited.
Input:
Explain zero-trust architecture for AI workloads in 3 paragraphs. Include specific technical controls.Expected Output:
- Comparison view shows responses from 2–3 models side by side
- Each model's response quality and latency are visible
- PII protection applied identically across all models
Talking Points:
"Same prompt, multiple models, identical protection. Whether your team prefers GPT-4, Claude, or Gemini, GhostAI applies the same PII detection, the same tokenization, the same audit trail. No model gets special treatment — and no model gets raw data."
Chapter 5: Enterprise Controls
Duration: 3 minutes Story: Walk through the admin console: audit log showing every request and PII detection event, SIEM integration for forwarding events to your SOC, and compliance dashboard mapping controls to frameworks.
Input:
(No chat input — this chapter navigates admin panels)Expected Output:
- Audit Log: timestamped entries for every PII detection, memory operation, and model call
- SIEM: configured webhook/syslog destinations, event forwarding status
- Compliance Dashboard: framework-mapped controls (SOC 2, HIPAA, PCI, FedRAMP)
Talking Points:
"Every request is logged. Every PII detection is auditable. Every model call is traced from input to output. And all of it flows to your SIEM in real time. This isn't a checkbox — it's a living audit trail your compliance team can query right now."
Chapter 6: Industry Verticals
Duration: 3 minutes Story: Switch to the Healthcare vertical and process a clinical note with PHI. GhostAI applies healthcare-specific detection rules — MRN, diagnoses, medications, provider names — on top of base PII protection.
Input:
Summarize this clinical note for the patient chart:
Patient: Margaret Chen, DOB 11/23/1954, MRN 2847103. Presented to ED with acute chest pain radiating to left arm. BP 168/94, HR 102, SpO2 96%. ECG shows ST elevation in leads II, III, aVF. Troponin I elevated at 2.4 ng/mL. Started on heparin drip, aspirin 325mg, clopidogrel 600mg loading dose. Cardiology consulted — Dr. James Wilson recommends emergent cardiac catheterization. Patient consented. Next of kin Sarah Chen (daughter) notified at (847) 555-0142.Expected Output:
- Healthcare PHI detection badges: patient name, DOB, MRN, provider name, phone
- Clinical terms preserved (medications, vitals, diagnosis)
- Summarized clinical note with PHI tokenized
Talking Points:
"Same platform, different rules. When we switch to Healthcare, GhostAI adds PHI-specific detectors — MRN, provider names, next-of-kin details — on top of the base PII engine. The clinical content stays intact because your physicians need it. The identifiers vanish because HIPAA demands it."
Chapter 7: Geographic Intelligence
Duration: 3 minutes Story: GhostAI's map tools let the AI search for places, get directions, and display interactive maps inline. With the optional Google Maps Premium provider, the AI gains place details, distance matrices, transit directions, and satellite imagery — all audit-logged and controlled by admin policy.
Part A — Place Search (30 sec)
Input:
Find coffee shops near Times SquareExpected Output:
- AI calls
map_search— interactive map appears inline with numbered markers - Results show names, addresses, and categories
- Map renders in dark theme with style cycler (dark / light / terrain / satellite / hybrid)
Talking Points:
"The AI detected geographic intent and called our map search tool. This map is rendered client-side with MapLibre GL — the same engine whether you're on our default privacy-first OSM stack or Google Maps Premium."
Part B — Place Details: The Hero Moment (1 min)
Input:
Tell me more about the Blue Bottle Coffee on Mint Plaza in San Francisco — what are the hours and reviews?Expected Output:
- AI calls
map_place_details— a rich card renders inline with star rating, review count, opening hours, phone, website, and photos - "Powered by Google" attribution badge appears
Talking Points:
"This is something our default OSM provider can't do. With Google Maps Premium, the AI calls the Places API to get ratings, hours, photos, and reviews — all rendered inline. And every one of these API calls is audit-logged. Watch —"
(Navigate to Platform Admin → AI Audit, filter for
GEO_EXTERNAL_QUERY)"There it is. Timestamped, with the endpoint and query type. Your security team has full visibility into every external geo query."
Part C — Distance Matrix (30 sec)
Input:
How long does it take to drive from 350 5th Avenue in Manhattan to JFK Airport, LaGuardia Airport, and Newark Airport?Expected Output:
- AI calls
map_distance_matrix— a clean table renders with distance and duration for each origin/destination pair - Three rows: JFK (~45 min), LaGuardia (~25 min), Newark (~35 min)
Talking Points:
"One question, three comparisons. The distance matrix API handles N-to-N calculations in a single call — perfect for logistics, real estate, or any multi-location planning."
Part D — Admin Controls & the Privacy Story (1 min)
Input:
(No chat input — navigate to Platform Admin → Geo Provider)Expected Output:
- Geo Provider admin panel showing: provider set to
hybrid, masked API key, four capability toggles, amber privacy notice - Current status panel showing active capabilities
Talking Points:
"Here's the control plane. The provider is set to Hybrid — that means Google handles search and place lookups, but all routing stays on our self-hosted OSRM. Your route data never leaves your infrastructure."
(Point to the privacy notice)
"Every Google API call is audit-logged. The API key is encrypted at rest with AES-256-GCM. And if Google goes down? The circuit breaker kicks in and everything falls back to OSM automatically — your users never see an error."
(Toggle provider to
osm, save)"One toggle. Now we're back to fully self-hosted, zero-external-call mode. No code changes, no redeployment. That's the promise: enterprise geo capabilities when you need them, privacy-first by default."
(Toggle back to
hybridfor the rest of the demo)
Optional — GDPR Guard (for EU-focused audiences):
(Set org data region to
eu-west-1in admin, attempt a Google-powered query)"For GDPR-region organizations, the Google provider is blocked until an admin explicitly acknowledges cross-border data transfer. You can't accidentally send location queries to Google from an EU tenant. That's compliance enforced in code, not just in policy."