FLAGSHIP · TEACHER AI
The conversational
voice agent architect.
Building a voice agent means juggling tone, timing, tools, and dozens of other knobs. Teacher AI takes the brief in plain language and ships a production-ready agent—like an expert pair-programmer who already knows voice.
01 · Generate
Describe the goal. Get a production agent.
Plain English or Hinglish is enough. Teacher AI injects industry best practices, asks up to three clarifying questions when things are fuzzy, then ships a hardened prompt with state machines, objections, and safety rails.
GENERATE
Hearing goal“Make a polite collections agent that handles EMI disputes…”
Verify identity via DOB or OTP?
OTP before loan talk
Allow partial EMI settlement?
Yes — min ₹500 plan
WhatsApp link when they agree?
send_whatsapp_link
Prompt · state machine · objections · guardrails
02 · Modify
Surgical edits. Nothing else gets erased.
Ask for one change—more empathy, a WhatsApp link, a softer EMI path—and Teacher AI patches only that behavior. Existing rules, variables, and edge cases stay intact.
MODIFY
Scanning prompt“More polite · handle disputes · send WhatsApp if they agree”
Opening tone
Warm greeting kept
EMI dispute path
Scanning…
WhatsApp payment tool
Scanning…
Identity check
OTP gate unchanged
Transfer rules
Escalation intact
Only the requested behavior changes
03 · Full stack
Mind and senses, tuned together.
Not just a prompt. Teacher AI configures Brain (prompt, variables, tools, KB), Senses (STT/TTS), Experience (eagerness, fillers, ambience), and Post-call BI (dispositions, entities, goals) in one pass.
FULL STACK
RecommendingOne recommendation.
Four pillars tuned.
Brain
Priya · {{emi_amount}} · transfer_call
Senses
hi-IN · conversational · 1.05×
Experience
Balanced · Haanji · soft office
Post-call
PTP · Dispute · containment
Mind and senses, configured together
04 · Diff & Checkpoint
See every change. Restore any turn.
Git-style diffs show exactly what was added or removed before you apply. Every conversational turn saves a checkpoint—roll back prompt and full-stack config if an experiment goes sideways.
DIFF & CHECKPOINT
+2 · −1 pendingCHANGES
+2 · −1
- Demand immediate UPI payment.
+ Offer empathy + minimum ₹500 plan.
+ Call send_whatsapp_link on agree.
TIMELINE
Turn 1
01Polite opening
Turn 2
02EMI dispute path
Turn 3
03Restored · WhatsApp tool
See every change. Restore any turn.
05 · KB & Tools
Need a policy PDF? Attach it in the flow.
When Teacher AI spots a gap—cancellation rules, a doctor roster, a rate sheet—it asks you to upload files or paste a URL right in the conversation. It binds tools like Sheets sync, calendar booking, and WhatsApp, then writes when-to-call guidance into the prompt.
KB & TOOLS
Gap detected“You’ll need the cancellation policy and doctor roster for this clinic agent…”
cancellation-policy.pdf
Knowledge base · PDF
clinic.vaani.ai/roster
Knowledge base · URL
Sheets sync
Log outcome after each call
Calendar booking
When they pick a slot
WhatsApp link
On payment agreement
Attached in-flow · guidance written into the prompt
NEXT
Days of prompt drafting → minutes of conversation.
Generate, modify, configure the full stack, and roll back safely— then ship.