AHM Operator
Workflow on top — drag to reorder. Voice / prompt / models below.
Workflow (interview script)
3 main 0 followups- 1main
Sekarang {{candidate_first_name}} menetap di mana? apakah di Jakarta atau di luar Jakarta?
order #0 - 2main
Dari CVnya, sebelumnya sebagai {{cv_last_role}} di {{cv_last_employer}}, itu ngapain aja?
order #1 - 3main
kemampuan kamu di CV: {{cv_skills}} banyak, berarti udah banyak pengalaman, base Salary yang kamu harapkan di angka berapa?
order #2
First message (greeting)
212 charsSystem prompt
700 charsThe workflow's main questions get appended to this prompt at dispatch time. KB context is fetched live by the agent — don't hard-code it here.
Session settings
Workflow has 3 main questions — agent uses that count. Stored config value (10) is ignored while workflow is set.
Kandidat harus grant izin kamera di device check; kalau decline, dia nggak bisa lanjut. Cocok untuk posisi customer-facing / hiring real (per AHM HR — verify identity).
Subtitle live muncul di bawah avatar — pertanyaan AI dan ucapan kandidat yang baru di-transcribe. Bantu kandidat yang gak fasih bahasa atau lingkungan ribut. Transcript tetap kerekam buat grading walau toggle off.
Interviewer avatar
Mood
Models
Pulled from lib/agent-catalog.ts — the dashboard's allowlist of providers the agent has plugins for. Adding a new model here without installing the matching livekit-plugins-* in the agent will crash dispatch.
Language model (LLM)
Default. Stronger reasoning than 2.5 with comparable latency at thinking_level=low. Global endpoint only (no asia-southeast1).
Voice (TTS)
Lowest latency; slightly less natural.
You can paste a custom voice id (e.g. an ElevenLabs cloned voice) into a future input — it stays saved even if not in this list.
Eleven applies VoiceSettings.speed in [0.8 – 1.2]; values outside that range are clamped at dispatch.
Lower values let the voice sound livelier; higher values keep tone tightly consistent across the call.
How closely the output should match the source voice. Too high can amplify any noise in the original sample.
Region only applies to Google Chirp 3. ElevenLabs runs from ElevenLabs' own infrastructure and ignores this.
Transcription (STT)
Press Enter or comma to add.
Press Enter or comma to add.
Custom post-processing prompt
2997 chars · okTemplates (klik untuk pakai)
Variabel tersedia: {{transcript}}, {{candidate_name}}, {{language}}, {{salary_budget_min}}, {{salary_budget_max}}, {{salary_currency}}, {{preferred_locations}}, {{psychology_guidance}}. The system enforces strict JSON; markdown / preamble in the response will fail to parse.