Train the MultiplyMii Founder Voice Guide toward fast, valuable answers.
This page is a live test environment for Ofri, Yoni, and the team. Ofri can test the agent directly, give feedback during the call, and also log structured notes below. The goal is to make the agent more direct, more useful, and more outcome-oriented while staying positive and in Yoni’s founder/operator style.
The agent is live but this page is hidden/noindex and not linked publicly. Feedback submitted here stays in your browser until copied or downloaded.
Answer first. Ask less. Guide faster.
The agent should answer the caller’s question quickly, ask the minimum high-value questions needed, give one useful founder/operator read, then route toward the right next step.
Rhythm: Answer → one clarifying question → useful read → route.
Avoid: long monologues, serial questioning, vague qualification, or pushing a call too early.
Route: discovery call if unclear; sourcing kickoff if the role, owner, timeline, and outcomes are already clear.
Use the floating “Start a call” bubble.
Run one scenario at a time. Ofri can speak feedback out loud during a test call — for example, “this is Ofri giving calibration feedback” — and should still capture the exact moment below so Hermes/Yoni can review and decide what becomes a real patch.
What we are tuning now
- Shorter opening: role ideas, KPIs, indicative all-in pricing, and model guidance up front.
- Approved 2026 all-in cost cheat sheet loaded for common roles.
- Blended-role requests now trigger focus-first guidance, not sourcing kickoff.
- Timeline answer now anchors on first three candidates in about nine days.
- Project-work boundary and part-time pricing language updated from Yoni feedback.
- Tester feedback is calibration-only; public callers cannot teach or change the agent.
- Self-assessment monitor now reviews new visible ElevenLabs test calls.
- Pronounce MultiplyMii as “Multiply Me”; written brand stays MultiplyMii.
- Max two questions before a “my read is…” synthesis.
- Earlier classification: discovery, sourcing kickoff, not-yet-fit, or human boundary.
- Faster Yoni-style voice cadence.
- Positive Yoni-style directness, not generic chatbot politeness.
You can give feedback to the agent while testing.
Use the live call like a calibration session. If the agent says something too vague, too long, too generic, or off-model, say the correction out loud and then log it below.
In-call feedback: say “this is Ofri giving calibration feedback” before the note, so the recording is easy to review.
Structured note: use the feedback form to capture the caller prompt, agent answer, issue, desired behavior, and ideal wording.
Important boundary: the agent does not train itself live. Feedback is reviewed by Hermes/Yoni and only approved changes become prompt, knowledge, CTA, or boundary patches.
Test the paths that matter.
Turn testing into training data.
Ofri’s feedback should become a prompt, knowledge, CTA, or boundary change. Use exact examples wherever possible.
Copy this back to Hermes/Yoni
Pick a scenario, run the agent, then generate a feedback packet.
No feedback is sent automatically. Ofri can speak feedback during the call, but durable changes only happen after Hermes/Yoni review the recording or feedback packet.
Every useful note becomes one of six decisions.
Prompt patch: change how the agent behaves, e.g. answer-first, fewer questions, shorter turns.
Knowledge patch: add missing MultiplyMii context, examples, or approved explanations.
CTA patch: improve discovery vs sourcing-kickoff routing and booking language.
Boundary patch: tighten pricing, EOR/legal, candidate, client-example, or “not Yoni live” handling.
Voice patch: make it more direct, positive, concise, and Yoni-like without importing tangents.
Human decision: hold items that need Yoni, Ofri, Cat, or Lipi before changing the agent.