Blueprints [📲𝙋𝙧𝙤𝙙𝙪𝙘𝙩]

Added by Justin Sheehan Justin S. May 29, 2026 12:45am
Column
Done July 15, 2026 2:34am
Assigned to
Troy Pastoral Troy P.
Due on
Tue, Jul 14
Notes
Ambassador AI · Answer Mode Technical Blueprint.pdf Download
Ambassador AI Modes.pdf Download
Ambassador AI · Advisor Mode Technical Blueprint.pdf Download
Ambassador AI · Ambassador Mode Technical Blueprint.pdf Download
Subtasks
Implement new blue prints - NOT advisor mode, but the three blueprints in modes Troy Pastoral Troy P. Tue, Jul 14
Troy Pastoral
Troy Pastoral AI Whisperer June 3, 2026 3:49am June 3, 2026 3:49am
Advisor Mode v1 — dodenai-ae-v1

What changed: The Advisor Mode in Ambassador AI now fully matches the v1.0.0 technical blueprint. When a user sends a problem or decision question, the AI routes it through an upgraded pipeline: smarter knowledge filtering, a 30-requests-per-minute cap, automatic promotion/demotion to the right mode, and a source list shown below every advisor response. Escalation now detects performance management and strategic commitment topics and flags them with elevated audit logging.

What to test:

1. Advisor source list renders below response
Before: No source list appeared after advisor responses.
After: A numbered source list ([1] Title — collection — updated date) appears below the AI response.
Steps: Open chat, switch to Advisor mode, ask a workplace question (e.g. "How should I handle a difficult performance review?"), check that a Sources section appears below the answer.

2. Advisor promotes to Ambassador on mission/values questions
Before: Mission-related questions stayed in Advisor mode silently.
After: A promotion signal is sent so the UI can prompt switching to Ambassador mode.
Steps: In Advisor mode, ask "What does our company stand for?" — the response header should include a mode promotion signal toward Ambassador.

3. Advisor demotes to Answer on simple lookup questions
Before: Quick-fact questions stayed in Advisor mode.
After: A demotion signal toward Answer mode is sent for simple lookups.
Steps: In Advisor mode, ask "What does OSHA stand for?" — demotion signal toward Answer mode should be emitted.

4. Escalation flags for performance management topics
Before: "performance review" or "PIP" questions had no special escalation notice.
After: A polite escalation notice appears suggesting the user loop in Manager and HR, with advisory disclaimer.
Steps: Ask "How do I put someone on a performance improvement plan?" — an escalation suggestion should appear at the bottom of the response.

5. Escalation flags for strategic commitment topics
Before: "long-term strategic commitment" questions had no escalation flag.
After: An escalation notice recommending Executive Sponsor consultation appears.
Steps: Ask "Should we commit to a 3-year partnership with this vendor?" — escalation notice should appear.

6. Knowledge not in the knowledge base -- caveat prefix
Before: If no matching policy docs were found, the AI answered without caveat.
After: Response starts with "I don't have policy specifics on this — here's what I'd think through, but please verify with the appropriate authority:"
Steps: Ask a very niche policy question unlikely to be in the KB; check response starts with the caveat.

7. Rate limit enforced at 30 per minute for advisor
Before: Advisor and other modes shared the same 60 req/min limit.
After: Advisor allows 30 requests per minute; 31st request in 60 seconds returns 429.
Steps: Technical test — rapid-fire 31 advisor requests in under 60 seconds; 31st should return "rate limit exceeded".
Troy Pastoral
Troy Pastoral AI Whisperer June 4, 2026 5:02am June 4, 2026 5:02am
Advisor Mode UI — Doden AI

What changed: Advisor mode was working in the backend but
had no visible way to switch or discover it. Added a
Coach/Advisor toggle in the chat header and a slash command
popup that appears when typing /.

What to test:

1. Coach / Advisor toggle in header
Before: No visible way to switch modes in active chat
After: Segmented pill (Daryle AI | Advisor) in header
Steps: Open chat. Look top-left next to yellow button.
Click "Advisor" — highlights. Click
"Daryle AI" — returns to coach mode.

2. Slash command popup
Before: Typing /advisor had no visual cue
After: Menu appears above input showing /advisor
Steps: Click input. Type /. Popup shows
"Advisor — Bring a problem." Click it
— input fills with /advisor ready to type.
Edge case: /adv still shows menu. Space after
/advisor hides it.

3. Escape dismisses
Steps: Type /. Menu appears. Press Escape. Gone.

4. Advisor placeholder
Before: Always "Ask me anything..."
After: Advisor mode: "What challenge or decision
do you need guidance on?"
Steps: Toggle to Advisor. Clear input. Placeholder
changes. Toggle back — resets.
Troy Pastoral
Troy Pastoral AI Whisperer June 4, 2026 5:03am June 4, 2026 5:03am

Will need QA for this cc Justin Red

Justin Sheehan
Justin Sheehan Chief Journey Officer June 4, 2026 5:17am June 4, 2026 5:17am

Troy sounds good. Jether Red


I will also have Ben and Malina test.

Troy Pastoral
Troy Pastoral AI Whisperer June 9, 2026 7:37am June 9, 2026 7:37am

cc Red

Troy Pastoral
Troy Pastoral AI Whisperer Edited July 14, 2026 10:34am July 14, 2026 10:34am
image.png
image.png Download

Answer, Advisor, Ambassador — QA checklist, plain English


Basics that apply to all 3


  • Chat header shows 3 mode buttons, right names, right icons
  • Switching modes mid-chat shows a clear divider, doesn't crash
  • Old chats from before this update still open and load fine


Answer mode (the default, general-purpose one)


  • Opens by default for new chats
  • Handles normal traffic without getting throttled
  • When it can't find an answer in the knowledge base, behaves same as before


Advisor mode


  • Recently fixed bug: it was declining to answer too often, even when it actually had good info. Confirm it now only declines when it truly has nothing relevant.
  • Ask it something it should know → gets a real answer, not a false "I don't know"
  • Ask it something outside its knowledge → politely declines, doesn't make things up


Ambassador mode (the mission-focused one)


  • Never invents answers about the mission — if it doesn't have a verified source, it declines and offers to redirect
  • Can't be hammered with requests — caps at 20 per minute
  • Tone and answer style match the approved script (calm, on-brand, short quotes)
  • Known unresolved issue — the content library isn't fully labeled yet, so the safety filter that's supposed to block off-topic material isn't fully active. Only the "how relevant is this" score is currently doing the filtering. Flag this as a known gap, not a passed test, until the labeling work is done.


    cc Red
Red Bañares
Red Bañares Astra Product Experience Designer July 15, 2026 2:35am July 15, 2026 2:35am

Noted on this Troy

I have updated the tester with these new modes. The assessment will check Daryle's answers based on these expected behaviors of each mode.

cc Justin Ann Brandon