Candor · concept demo by Hexaview Technologies, Inc. for Vantage Leadership Consulting· Candidates and client companies are fictional
Hexaview Technologies, Inc. · concept demo for Vantage Leadership Consulting · about 9 minutes
Candor
An AI workspace for Vantage's consultants and partners. It helps write evidence-backed assessment reports faster and runs helper agents alongside the team. It is built so that private information stays private, every statement links to its source, and a person makes every decision.
No sign-in, no setup. The candidates, client companies and numbers are fictional. Best on a laptop. New to this? Read on first, it takes about four minutes.
An agent working alongside the team
Two experts scored a candidate differently on leading change (4 vs 2). The agent shows the evidence each one used so they can agree.
What the AI actually receives
[PERSON_01], a senior operations leader, reorganized a two-shift schedule.
Safety test · a hidden instruction inside colleague feedback
Blocked before the AI acted on it. The comment is treated as text, not a command, and flagged for a person to review.
The background in 60 seconds
When a company picks a senior leader, it often asks experts to assess the candidates first
Vantage Leadership Consulting, based in Chicago, has helped organizations select and develop leaders for 50 years. Its consultants assess executives with interviews, psychometric tests, colleague feedback (called 360 feedback) and business simulations, then write a report the client uses to decide. In this demo you play Jordan Rivera, a Vantage partner who reviews and approves the work. Elena Carter and Sam Patel are the two consultants who interviewed and assessed the candidate. The client company, Northwind Manufacturing, and its candidates are made up.
1A company needs a leaderNorthwind needs a new VP of Operations and shortlists four candidates.
2Vantage gathers evidenceInterviews, tests, colleague feedback and a role-play exercise for each candidate.
3A consultant writes a reportStrengths, risks and development areas, measured against what the role needs.
4The company decidesThe report informs who gets the job. People's careers depend on it.
The problem
Step 3 is slow, and AI could speed it up, but the stakes are unusually high
It takes senior experts hours
Turning pages of notes, scores and comments into a careful report is skilled, manual work, done for every candidate.
The data is deeply private
Personality results and candid colleague feedback about named executives. One leak, or one client's data showing up in another client's work, and trust is gone.
The rules are tightening
In Illinois, where Vantage is based, employers must now tell people when AI is used in hiring or promotion decisions (since January 2026), and using AI in a way that discriminates can count as a civil rights violation. Vantage's clients will expect help to comply.
So why not just paste the notes into a chatbot?
What could go wrong
What Candor does instead
Private names and details leave the firm
Names are swapped for placeholders before the AI sees anything, and the most private option runs the AI on Vantage's own servers.
One client's information ends up in another client's answer
Every client's files sit in a separate locked vault. A request about another client is blocked before any search happens.
The AI states something nobody actually said
Every sentence links to the exact note, score or comment behind it. Where there is no evidence, it says so and writes nothing.
Someone tricks the AI with a clever message
Several layers of safety checks stop known tricks, and a built-in set of attack tests reruns after every change.
Biased wording slips into a report about a real person
A second AI reviewer flags age, gender and family-related wording before release, and a person decides on each flag.
What Candor does
Six promises Vantage can show its clients, not just tell them
NorthwindMeridian · locked
Every client is walled off
Separate storage and encryption for each client. People and AI only reach the client they are working on.
Dana Whitfield→[PERSON_01]
The AI never sees names
Names and identifying details are replaced first. Vantage can even run the AI on its own servers.
Calm under pressure2360 comments 03, 07, 11
Every sentence shows its proof
Click any sentence in a draft to see the source. Gaps in evidence are flagged instead of guessed.
Five agents, each named for a part of Vantage's work: assessment, bias audit, coaching, succession and knowledge reuse. Two more run in the background to mask personal data and record sources. They suggest; people decide.
ATTACK hidden instructionblocked · first check48 of 48 tests stopped
Safety checks you can watch
Clean Room Chat checks every message before the AI sees it: personal details masked as you type, hidden instructions removed, other clients' data refused. A 48-test suite, mapped to the OWASP list of AI security risks, reruns after every change.
✓Approved by a partner
A person signs off
The AI never rates a candidate or recommends a hire. Nothing reaches the client until a partner approves, and the client gets a plain AI-use notice.
All of it in the tools the team already uses
The assistant answers questions, drafts sections and prepares meetings, inside Candor and inside Microsoft Teams, where agents post findings and partners approve from a card.
What you'll see in the demo
One partner's morning, from the practice dashboard to a released report
Act 1 · The practice
The firm at a glance
Pipeline, turnaround and client health
Agents already at work
A morning brief
Act 2 · Trust
How data is protected
One locked vault per client
A candidate's protected file
Exactly what the AI receives
Act 3 · Agents at work
Help that runs in parallel
Findings from four agents
A live succession 9-box
Configuring agents and picking the AI model
Act 4 · The report
From draft to release
Cited draft, refused guesses
Bias review and partner approval
Act 5 · Clean Room Chat
Try to break it
Watch personal details masked as you type
Two trick attempts, stopped in front of you
The 48-test suite
Act 6 · Everywhere
Teams and coaching
Candor inside Microsoft Teams
An agent drafting coaching nudges
How to use the demo
A guide card walks you through it. You can't break anything.
⋮⋮ACT 5 · CLEAN ROOM CHAT⟲›–×
An instruction hidden in pasted textA candidate's note tells the AI to rate her "Strong"…▶ Run flow 2▸ For the build teamBackNext21 of 27
Next and Back move through 27 short steps. Arrow keys work too.
▶ buttons make something happen live, such as an agent finding something or a trick attempt being stopped.
The highlighted box on screen is what each step is talking about.
Drag the top bar to move the card. – shrinks it, › goes to the next step while it is small, and ⟲ puts it back where it was placed automatically.
For the build team opens a short technical note on each step. Skip it if you're not building it.
The assistant on the right answers what you type. When a step is about the chat, it opens wide so it is easy to read. Clean Room Chat, in the left menu, shows every message being checked and masked before the AI sees it. Try these:
Overview in the top bar brings you back to this page at any time.
What's real and what isn't
A clickable preview of a product we can build
Every screen, agent setting, attack and flow shows how the product will work.
The assistant's answers and the Clean Room results are scripted for this demo. In the live build they come from real models working on Vantage's own files.
Northwind, the other client companies, all candidates, scores, fees and dates are invented.
A working pilot with real AI, real vaults, agents and audit logs can follow quickly. We'll scope it with you.
Why Hexaview is building this
Privacy first, then speed
Firms that make sensitive decisions about people need AI they can defend to their clients. Candor puts the privacy controls first: the most private model option is the default, agents can only suggest, and every guardrail is tested in the open. It is built with HAKI AI, Hexaview's method for delivering AI projects, with human approval at every release.
Ready? It takes about nine minutes.
⌕ Search your engagements
JRJordan RiveraPartner
Friday, October 23 · Practice overview
Good morning, Jordan
Firm-wide view across every active engagement. These figures are totals only: no candidate detail crosses from one client to another.
Agents working alongside your team3 running · 41 findings this week, 33 accepted
Engagement pipeline
Engagements by stage
Notes-to-report turnaround
Average days, last 12 weeks
Client portfolio
Click a client to open it
Client and engagement
Stage
Health
Next milestone
Fees
NMNorthwind ManufacturingVP Operations assessment · 4 candidates · J. Rivera
Partner review
On track
Final reports due Nov 4
$86,000
LLLakeshore LogisticsCOO succession bench · 6 leaders · J. Rivera
Evidence collection
At risk
2 of 6 360s overdue · readout Dec 2
$120,000
BFBrightwater FoodsExecutive coaching · 2 leaders · J. Rivera
Active
On track
Marcus Hale, session 7 · Oct 28
$48,000
MHMeridian HealthCFO assessment · M. Okonkwo
Drafting
Summary only
Not your engagement. Candidate detail stays in Meridian's vault.
$92,000
HCHarbor Credit UnionBoard self-evaluation · 9 directors · S. Lindqvist
Scoping
On track
Kickoff Nov 2
$35,000
CECrestline EnergyHigh-performance team · J. Rivera
Released
Complete
Follow-up pulse Jan 12
$64,000
Fairness review
Items flagged and resolved per week
Succession bench · Lakeshore COO
This week
Northwind check-in with Priya Shah, CHROYou and Elena Carter
Follow-up interview: Dana WhitfieldSam Patel · commercial acumen
Coaching session 7: Marcus HaleYou · Brightwater Foods
Harbor Credit Union board kickoffS. Lindqvist
Northwind final reports due4 candidates
Consultant capacity
Booked hours this week
Where the hours went
Per released report, this quarter
Engagement
Northwind Manufacturing · VP Operations
🔒 Northwind vault · key nw-kms-01Lead partner: Jordan RiveraAssessors: Elena Carter, Sam PatelReport due Nov 4Retention: 24 months
Candidates
Candidate
Interviews
360
Evidence
Report
DW
2 of 2
14 of 14
86%
Partner review
RM
2 of 2
11 of 12
78%
Ready to draft
LO
1 of 2
6 of 12
41%
Withdrew Oct 19
TB
2 of 2
13 of 13
100%
Released Oct 14
Click a candidate name to open their file.
Northwind success profile v3
Agreed with Northwind's CHRO on Sep 18. Every report is written against these six competencies.
Prepared by Vantage Leadership Consulting for Northwind Manufacturing · VP Operations · Confidential
Executive summary
Not drafted yet. Ask the assistant to draft it from the evidence →
Operational leadership
Dana describes reorganizing a two-shift plant schedule after a sustained quality problem, and both assessors rated the account as specific and well sequenced.1Peers consistently describe her as calm under pressure and quick to bring the right specialists into a problem.2Her LeadScale profile shows high structure and composure, with moderate tolerance for ambiguity.3In the plant crisis simulation she set a containment plan within the first ten minutes and assigned clear owners for each action.4
Leading change
Dana rates herself notably higher on leading change than her raters do, and the two assessors also differed on this competency.5Raters describe strong execution of change once direction is set, with less evidence of building early support before announcing it.6
Commercial acumen
Insufficient evidenceThe assistant did not write this section
No interview answer, 360 comment or simulation note speaks directly to pricing, margin or customer economics. Writing a judgment here would be a guess.
Evidence
Click any underlined sentence to see the exact source behind it.
Ratings
Set by the assessors. The assistant shows evidence strength but never sets a rating.
Operational excellence4 sources
Builds teams3 sources
Leads changeAssessors differ
Commercial acumenNo evidence
Fairness review
A second reviewer reads every word before release
The review agent checks the draft against wording rules agreed with Vantage's psychologists. You decide on each item. 4 items open
Age-related wordingExecutive summary, paragraph 2 · from assessor notes
"…a young, hungry leader with plenty of runway…"
Suggested: "sets ambitious targets and asks for stretch assignments"7
Why: describes age rather than behavior. Age-related wording in a selection report can be read as age discrimination.
✓ Resolved by Jordan Rivera
Family statusDevelopment priorities · from interview notes
"…family commitments may limit her availability for travel…"
Suggested: remove. Not related to the success profile and not supported by any source.
Why: family status is not a job criterion. The role's travel requirement was discussed directly and Dana confirmed it.8
✓ Resolved by Jordan Rivera
Gender-coded descriptorLeading change · from one 360 comment
"…can come across as abrasive in meetings…"
Suggested: "two raters said she sometimes moves to decisions before others have finished raising concerns"9
Why: "abrasive" is applied more often to women in leadership feedback. The suggestion uses the specific behavior raters described.
Suggested: remove, or add evidence. No interview answer, rating or comment supports this.
Why: every statement in a Vantage report should be defensible from the evidence.
✓ Resolved by Jordan Rivera
Approve and release
Nothing reaches the client without a partner
You sign off here as the lead partner. Release also prepares the AI use notice for Northwind.
Release checklist
✓
Every sentence has a source
23 of 23 sentences cited, 9 sources
!
Fairness items resolved
0 of 4 resolved
✓
Ratings set by an assessor
3 rated, 1 marked not rated for lack of evidence
4
Partner approval
Waiting for you
JR
Jordan Rivera
Lead partner
AI use notice · NW-0007
To: Northwind Manufacturing, Human Resources
Released
Vantage Leadership Consulting used an AI tool to help prepare the assessment report for the VP Operations candidate. The tool organized interview notes, assessment results and feedback, and produced a first draft with a reference for every statement.
The AI tool did not make or recommend any hiring decision and did not assign ratings. A Vantage consultant reviewed every statement and a Vantage partner approved the final report.
Candidate information was protected so that the AI tool did not receive names or contact details. Questions can be directed to Jordan Rivera.
Wording to be confirmed with Vantage's and Northwind's counsel. A record of this notice is kept in the audit log.
Agent and model studio
Agents, and the model behind them
Each agent watches one part of the work, does the legwork in the background and hands its findings to a person. None of them can rate a candidate, approve a report or send anything on its own.
AI model for the Northwind engagement
Every agent above, the assistant and Clean Room Chat use this model for this engagement. Switching it changes nothing else.
Most privateLeast private
Option 1Gemma, hosted by VantageOpen-weights model running on Vantage's own servers or private cloud. Prompts and data never leave Vantage's network.Recommended
Option 2Amazon Bedrock in Vantage's AWSManaged models inside Vantage's AWS account and chosen region. Prompts are not used to train the models.
Option 3Claude, Anthropic APINot used for training by default. Zero data retention where contracted.
Option 4OpenAI APINot used for training by default. Retention terms per agreement.
Same workflow, guardrails and audit trail on every option. Provider terms must match Vantage's signed agreements.
Always-on guardrail agents
These can't be switched off. They protect every agent above and every chat.
Vantage Client Enclave & PII Tokenization Agentvantage-client-enclave-pii-agentAlways on. Secures each client's data, replaces personal details with placeholders and lets no personal data reach a foundation model.
Vantage Evidence Provenance & Audit Agentvantage-evidence-provenance-agentAlways on. Stores the source for every statement, the reasoning trail and the audit log.
Disclosure agentvantage-disclosure-agentAlways on. Writes the client's AI use notice when a partner releases a report.
Live agent activity
Updates as agents work
Trust center
Every step recorded. Every guardrail tested.
Attack testing, audit trail and data erasure, for Vantage's leadership, IT and clients' security and legal teams.
Red-team results Last full run: Oct 20, after the prompt update
Every attack type is mapped to the OWASP Top 10 for LLM Applications (2025). The suite reruns after any model, prompt or policy change. To try attacks yourself, use Clean Room Chat.
12 attack families × 4 variants
Attack family
OWASP
Tests
Stopped
Usually stopped at
Audit trail · Northwind engagement
Time
Who
Event
Detail
09:02:11
E. Carter (assessor)
ingest
6 files to nw-vault · names replaced
09:03:02
Client Enclave & PII Tokenization Agent
tokenize
11 personal details replaced · 0 sent to models
09:04:37
Confidential Assessment Agent
model_call
Gemma (Vantage-hosted) · 3,812 tokens · draft v3
09:05:02
820 ILCS 42 & EEOC Bias Auditor
flags
4 raised on draft v3
09:07:40
Confidential Assessment Agent
finding
assessor ratings differ by 2 on Leads change
09:09:15
Confidential Assessment Agent
finding
self vs others gap −1.2 on Leads change
Forget a person
Removes everything about one assessee, under the retention policy, and keeps a receipt without personal data.
Source files6 stored
Search index entries214 stored
Placeholder map entries11 stored
Logs containing content38 stored
Laura withdrew from the process on Oct 19. Backups expire under the documented backup schedule.
Access policy
P-01 Client isolationEnforced
P-02 Names never sent to modelsEnforced
P-03 Partner approval before releaseEnforced
P-04 No hiring recommendations by AIEnforced
P-05 Agents suggest, people decideEnforced
P-06 No voice or video analysisEnforced
Blocked attempts this week: 0
Clean Room Chat
Every question is checked before the AI sees it
Type anything, or run one of the three flows. The right-hand panel shows, as you type, what the AI will actually receive and which checks your message passes.
Clean Room ChatNorthwind engagement · Gemma, hosted by Vantage
Try a flow:
Pick a flow above, or type your own message below. Try including a name, an email or a phone number.
More to try:
Live pre-filterupdates as you type
What you typed
Nothing yet
What the AI will receive
Nothing yet
0 personal details masked
Checks on send
Every message and check is written to the audit trail. Attempts that are stopped count as blocked attempts.
Candor in Microsoft Teams
The same assistant and agents, inside Teams
Consultants ask Candor in a channel, agents post their findings there, and partners approve from a card. Same vault scope, same guardrails, same audit trail. Type in the box below to try it.
⋮⋮⋮
⌕ Search
JR
Activity3
Chat
Teams
Calendar
Calls
Files
Candor
Teams
VAVantage Assessment
General
# Northwind VP Ops
# Lakeshore COO bench
Calibration room
VCVantage Coaching
General
Brightwater Foods
VANorthwind VP OpsMeet
PostsFilesCandor tracker+
CandorAPP8:30 AM
Good morning. Here's where the Northwind VP Operations search stands.
Northwind · VP Operations
Due Nov 4 · Northwind vault only
Tom Brennan
Released Oct 14
Dana Whitfield
Partner review · 4 fairness items open
Ravi Menon
Ready to draft · 78% evidence
Laura Okafor
Withdrew Oct 19
Vantage Confidential Assessment AgentAGENT9:07 AM
@Elena Carter@Sam Patel your ratings for Dana Whitfield differ by 2 points on Leads change.
Calibration suggested
Carter
4 · plant schedule change (interview p.4)
Patel
2 · announced plan before hearing shift leads (simulation, min 12)
SP
Sam Patel9:12 AM
Fair. I'll bring the simulation notes.
↩ Reply
JR
Jordan Rivera9:15 AM
@Candor how does Dana stack up against the CFO finalists at Meridian?
CandorAPP9:15 AM
Blocked by policy P-01, client isolation
This channel is scoped to Northwind. Nothing from Meridian was searched or sent anywhere. The attempt is logged for the firm admin.
↩ Reply
CandorAPP9:40 AM
@Jordan Rivera a report is waiting for your approval.
Approval needed: Dana Whitfield, assessment report v3
Citations
23 of 23 sentences sourced
Fairness
4 items open
Client notice
NW-0007, prepared on release
Aa ☺ 📎
Illustration of Candor's Teams app. Microsoft Teams is a trademark of Microsoft.
Coaching journey
Marcus Hale · Brightwater Foods
VP Supply Chain · coached by you since August · 6 of 10 sessions · Coachee consent on fileMarcus can see everything here
Development goals · from his assessment
Bring peers in before decisions are finalOn track
Pulse check from 5 peers: 3.1 → 3.8 since August
Delegate the weekly operations reviewNeeds attention
Still chairing 4 of 4 reviews in October
Present supply risk to the boardScheduled
Board session Nov 18 · rehearsal booked Nov 11
Between sessions
Oct 14 · Session 6
Practiced a pre-decision check-in script
Oct 16 · Nudge sent
"Before Thursday's review, who could chair it instead?"
Oct 20 · Marcus replied
Asked his deputy to chair next week
Oct 28 · Session 7
Assistant prepares session notes the day before
⋮⋮
For the build team
Drag the bar to move this card. Arrow keys also work.
Forget Laura Okafor?
This removes her files, search index entries, placeholder map and content logs from the Northwind vault. It cannot be undone. A receipt without personal data is kept.