Model Panel [📲𝙋𝙧𝙤𝙙𝙪𝙘𝙩]

Added by Justin Sheehan Justin S. August 27, 2026 3:59am
Column
Review
Assigned to
Troy Pastoral Troy P.
Due on
Fri, Aug 28
Notes

Daryle AI Model Comparison Brief


Overview


We want to add a model comparison experience within Daryle AI that allows a user to submit one prompt and see how multiple AI models respond to the same question.


The purpose is to demonstrate what makes Daryle AI and the Ambassador accumulated knowledge unique compared with a standard AI model. The experience should first show each model’s raw response independently, allowing the user to compare the different perspectives before receiving a combined analysis.


How It Should Work


  1. The user enters one question or prompt.
  2. Daryle AI sends that same prompt to multiple AI models.
  3. Each model generates its own independent response.
  4. The user can choose how to view the responses:

    • In a split-screen comparison showing all three responses at the same time.
    • In a group-chat-style view where the models respond one after another in the same conversation.
  5. Each response should clearly identify which model generated it.
  6. After all raw responses are displayed, Daryle AI generates a summary and analysis of the responses.


The initial group-chat or comparison experience should include:


  • ChatGPT
  • Claude
  • Gemini


Daryle AI / Ambassador accumulated knowledge may be included as an additional model or layer, depending on the final product design.


Model Comparison Views


Split-Screen View


The user enters one prompt and receives responses from all three models in separate sections of the screen. This allows the user to quickly compare the answers side by side.


One Prompt → Three Responses → Easy Comparison → Daryle AI Analysis


Group-Chat View


The user enters one prompt, and the models respond one after another, almost like participants in a group chat.


The conversation could appear as:


User:
“How should we think about developing a new leader within an organization?”


ChatGPT:
Response from ChatGPT.


Claude:
Response from Claude.


Gemini:
Response from Gemini.


Daryle AI Analysis:
Summary and analysis of the responses.


This format makes the experience feel conversational while still allowing the user to see how each model approaches the same question.


Raw Responses First, Analysis Second


The order of the experience is important.


Daryle AI should first display the complete raw response from each model without combining, editing, or summarizing the answers. This allows the user to see each model’s independent perspective and evaluate the differences directly.


Only after the raw responses have been presented should Daryle AI provide a combined summary or analysis.


The analysis could include:


  • A summary of the main ideas shared across the responses.
  • Important differences between the models.
  • Strengths and weaknesses of each response.
  • Areas where the models agree or disagree.
  • Missing considerations or potential blind spots.
  • A recommended or synthesized answer.
  • A perspective informed by Ambassador’s accumulated knowledge, principles, language, and frameworks.


The experience should clearly distinguish between:


  1. What each model said
  2. How the responses compare
  3. What Daryle AI concludes or recommends


Model Council-Style Analysis


Similar to Perplexity’s Model Council, Daryle AI should act as a final reviewing layer after the individual models have responded.


The analysis should not simply repeat the answers. It should evaluate them and help the user understand the significance of the different perspectives.


For example, Daryle AI could say:


“All three models emphasize the importance of mentoring and responsibility. ChatGPT focuses primarily on leadership development practices, Claude emphasizes relational trust, and Gemini highlights structured accountability. From an Ambassador perspective, the most important additional consideration is whether the emerging leader is being developed in alignment with the organization’s values and long-term purpose.”


This final analysis is where Daryle AI can demonstrate its unique value. It should help the user move from multiple answers to greater clarity, insight, and action.


What We Want to Demonstrate


The goal is not simply to prove that different AI models produce different wording. We want the user to clearly see how Daryle AI brings a different perspective because it understands Ambassador’s accumulated knowledge, principles, language, and way of thinking.


For example:


Prompt:
“How should we think about developing a new leader within an organization?”


The user could then compare:


Model

Response

ChatGPT

Standard leadership perspective

Claude

Alternative leadership perspective

Gemini

Alternative or deeper reasoning

Daryle AI Analysis

Comparison, synthesis, and recommendations informed by Ambassador knowledge and Daryle’s frameworks


Key Product Requirements


  • Display the raw response from each model before providing any summary or analysis.
  • Clearly label every response by model.
  • Do not alter or combine the raw responses.
  • Provide a final Daryle AI summary and analysis after all model responses are complete.
  • Clearly distinguish raw model responses from Daryle AI’s interpretation.
  • Allow users to switch between split-screen and group-chat-style views if both are included in the first version.
  • Make the analysis useful, not merely repetitive.
  • Use Ambassador’s accumulated knowledge to add context, identify gaps, and provide a distinct perspective.


The experience should feel simple:


One Prompt → Multiple Raw Responses → Easy Comparison → Daryle AI Summary and Analysis


Longer-Term Opportunity


This model comparison and analysis capability supports a larger vision for Daryle AI. Instead of requiring people to abandon the AI tools they already use, we can make Ambassador’s accumulated knowledge available alongside those tools.


The model comparison becomes a tangible way to show that the value is not simply another AI model. The value is the unique knowledge, perspective, and analysis Daryle AI brings to the conversation.

Subtasks
Implementation of requested feature Troy P. Sun, Aug 30
QA Testing Red B. Sun, Aug 30
QA Testing Ann L. Sat, Aug 29
Ann Lim August 27, 2026 1:59pm August 27, 2026 1:59pm

Justin question on the requirements:


I think it will be required to have a login for the 3 chatbots. For Gemini, what account should we use? And for Claude and ChatGPT, I assume we are using ambassador's?

Justin Sheehan Chief Journey Officer August 27, 2026 2:06pm August 27, 2026 2:06pm

Ann


This is built into Daryle.AI, so it should be pulling from the APIs. This is on and inside the Daryle.AI platform. Not separate accounts or individual logins.


Troy

Troy Pastoral AI Whisperer August 27, 2026 2:07pm August 27, 2026 2:07pm

Yep, we're using ambassador for this.


I'll pushing a change once testing is done

Troy Pastoral AI Whisperer September 10, 2026 9:38am September 10, 2026 9:38am
  1. Split-screen view
    Before: n/a (new feature)
    After: prompt sends to ChatGPT, Claude,
    Gemini at once, all 3 shown side by side
    Steps: enter one prompt, pick split view,
    confirm all 3 answer independently and
    labeled by model
  2. Group-chat view
    Before: n/a
    After: same 3 responses shown one after
    another, chat style
    Steps: toggle to group view, same prompt,
    confirm no reload needed, all 3 still
    shown
  3. Raw responses render clean
    Before: markdown showed as literal text
    (### 1., bold visible as symbols)
    After: renders as real bold/headers/lists
    Steps: check any response with
    formatting, confirm no stray symbols
  4. Ambassador-grounded analysis
    Before: analysis said "no sources" / zero
    citations, even on leadership/mission
    prompts
    After: cites real Ambassador documents
    when relevant
    Steps: ask "How should we think about
    developing a new leader within an
    organization?" — confirm analysis section
    shows cited sources, not empty
    Edge case: generic factual prompts (e.g.
    "capital of France") correctly show zero
    citations — that's expected, not a bug
  5. Order — raw first, analysis after
    Before: n/a
    After: all 3 raw answers finish before
    Daryle's analysis appears
    Steps: watch load order, analysis should
    Edge case: generic factual prompts (e.g. "capital of France")
    correctly show zero citations — that's expected, not a bug
  6. Order — raw first, analysis after
    Before: n/a
    After: all 3 raw answers finish before Daryle's analysis appears
    Steps: watch load order, analysis should never appear first or
    mixed in
  7. One model fails, others still work
    Before: n/a
    After: if one provider errors, other 2 + analysis still complete
    Steps: hard to force manually — spot check only, note if any
    provider looks stuck/frozen
  8. Reload keeps state
    Before: n/a
    After: refreshing page keeps last view (split/group) and
    responses, not blank
    Steps: run a comparison, refresh page, confirm it's still there
  9. Findable without typing "/"
    Before: only reachable by typing "/" in chat
    After: unconfirmed — check if it's in the mode dropdown/menu now
    Steps: open a new chat, look for Compare/Model Panel option
    without typing anything