A twelve-minute walkthrough.
Switch users in the left menu as you go. Each step says what to show and what it proves.
Before you present
- Stop the app, then reset to a clean start:
npm run demo:reset - Start it again:
npm run dev, then openhttp://localhost:3000. On the live demo, visitors are read-only guests until they sign in with the access code: sign in before you present, so you can switch between the four people. - Open the Trust centre once to confirm the audit chain reads “Intact”.
- 1
The AI handles the routine
Auditor“Half of everything coming in is already dealt with. This person only sees what needs them.”
- Point at the “Handled by AI” tile and the review queue.
- Ask the assistant: “What needs my review?”
- 2
Answers that cite their source
Auditor“It answers from approved documents, shows the exact section, and refuses when it has no grounds.”
- Ask: “What does the SOP say about temperature excursions?” Click a source chip.
- Ask: “What does the SOP say about employee holiday pay?” It refuses rather than guess.
- 3
Why it was flagged, and the decision
Auditor“The AI proposes the factors. Fixed rules compute the grade and decide who looks. The auditor decides, with a reason that goes in the record.”
- Open the label reconciliation item (grade D4).
- Show the factor breakdown, the 75% confidence line and “grounded in the controlled documents”.
- Escalate it with a reason.
- 4
Sign-off with a recorded signature
QA head“Escalations go to a second person, never the same one. The signature records who, what it means and when.”
- Switch to Priya Raman (QA head) in the left menu.
- Open the escalation, choose the meaning, type the name to sign.
- Note that a risk level 4 item now waits for an executive.
- 5
The most serious risks need a second signature
Executive“Risk level 4 (RPN 24 or more) needs the QA head and then an executive. The executive cannot sign first, and nobody signs twice.”
- Switch to Elena Vasquez (executive).
- Sign as executive, then open the evidence pack and show both signatures.
- 6
The AI is checked before it is used
Anyone“Every AI proposal and every answer goes through fixed, versioned rules first. A failure stops it. Here is what those rules catch.”
- Show the rules and the rule-set version.
- In the Assurance lab, show a bad AI output being caught by a named rule.
- Say clearly that the regulation links are proposals awaiting review by a qualified professional.
- 7
Documents have versions, and only the current one is quoted
Auditor“Old procedures are kept as history, marked superseded, and never cited as current. One approved for next year is not used early.”
- Open the deviation SOP and its version history.
- Open the old version to show the Superseded banner.
- Ask the assistant how many working days an impact assessment takes: five, from the current version.
- 8
How risk is scored, and what each score demands
Anyone“Two ICH Q9 ideas are fixed rules: risk is scored on defined scales, and formality must match the risk.”
- Show the A1 to D4 grid with live counts.
- Point at a box marked “below required”: it always goes to a person.
- Open “How a record is handled” to show the whole five-step path.
- 9
Lowering a risk takes a named measure and a signature
Auditor“A risk goes down by doing something that is recorded, not by typing a smaller number. Accepting what remains is an electronic signature.”
- As the auditor, apply a measure to an item and describe what was done.
- Switch to the QA head and sign the residual risk.
- Open the item’s report to show the measure and the signature.
- 10
Where the AI needs a person who knows
Team member“Missing information is asked of a named team. The item stays blocked until they answer.”
- Switch to Sam Okafor (team member).
- Answer the supplier question.
- Switch back to the auditor and show the item return to the queue.
- 11
Records arriving live
Auditor“This is what runs all day: records flow in from the quality systems and are assessed as they land.”
- Switch the “Live feed” toggle on the console and watch records arrive and route.
- 12
How it fits the quality process
Anyone“Every record is placed on the defined process, tied to company controls, and mapped to the regulations. Gaps are shown, not hidden.”
- Process map: stages and what needs a person.
- Controls and validation: the regulation gap, and IQ / OQ / PQ status linked to live items.
- 13
Can it be trusted, and is it worth it
Anyone“Evidence the AI agrees with experts, with its misses shown, and a value model anyone can adjust.”
- Say clearly the data is illustrative until the pilot.
- Move the calculator’s sliders to match their team.
- 14
Nothing leaves, and you can check
Anyone“Every AI call is listed. The safeguards are verified against the running system, not just described.”
- Show the live chain verification and the audit trail export.
What is real
- The audit trail: hash-chained, append-only, enforced by the database
- The ICH Q9 grading and the routing rules
- The Assurance Layer: fixed, versioned checks on every AI proposal and answer
- Two signatures on risk level 4, and signed residual-risk acceptance
- Document versions and dates: only the version in force is quoted
- Roles and separation of duties, enforced by the server
- Auditor decisions, team input, sign-off, evidence pack
- Grounding in documents: the exact section is cited, and it refuses without grounds
- Every action and every question recorded
What is simulated
- The AI’s proposals and the assistant’s wording
- The six system connections and all the records
- The SOPs and company controls (fictional)
- Login: four demo users stand in for real sign-in
- The mapping of each rule to a regulation: proposed, not yet reviewed by a qualified professional
- The e-signature’s identity check
- The accuracy figures on the Proof page: illustrative
Questions you will get
- Is the AI real?
- Not yet. It is a placeholder behind a single interface so a real, in-environment model replaces it without changing anything else. What you see is the workflow, the rules, the records and the controls around it.
- What stops the AI clearing something dangerous?
- It does not decide. Fixed rules route by risk, formality and confidence, and the AI can only add reasons for a person to look, never remove them. Before any of that, the Assurance Layer checks what the AI produced. The Proof page shows the misses too.
- What stops it making things up?
- It can only quote documents in force today, with the exact section, and every answer is checked before you see it: sources cited, versions current, quotations word for word, records real, no outside links. A failing answer is withheld.
- Is this validated or compliant?
- No, and the product says so. The checks and signatures are real, but this demo uses fictional data and a simulated AI, and the links from our rules to specific regulations are proposals that a qualified GxP professional has yet to review.
- Where does our data go?
- Nowhere. The design keeps regulated data and the model inside the customer’s own environment. The Trust centre lists every AI call.
- How long to a real pilot?
- See the delivery plan. A lean team reaches a shadow pilot in roughly 6 to 7 months.