07 · IVR Reference and Rail Facts

info · Part A is the content of IVR.pdf, transcribed from the pages. Part B is from Gnani's quick-start.md. Part C is research notes from earlier in our chat (my work, not yours). Treat Part C as "to verify with the rail owners", not as confirmed fact.

Part A · IVR.pdf (as in the document)

IVR

The automated Interactive Voice Response (IVR) call—or "flash call"—has become the gold standard for last-mile delivery operations in India. It is highly effective because it treats delivery as a communication problem rather than a mapping problem.

Instead of forcing a customer to stare at a moving pin, the system uses a smart cloud-telephony trigger to bridge the gap between the driver and the user.

How the Tech Works Behind the Scenes

The entire process is automated through the delivery agent's app and a cloud telephony platform (like Twilio, Exotel, or Knowlarity), ensuring the driver never manually dials your number while driving.

[Driver Marks Previous Stop Done]
              │
              ▼
[System Geofences Next Address] ──► [Cloud Telephony Triggers IVR] ──► [Customer Phone Rings]
        │
   ┌────┴───────────────────────┬───────────────────────┐
   ▼                            ▼                       ▼
[Press 1: I am home]    [Press 2: Leave at Gate]   [Press 3: Reschedule]
   │                            │                       │
Driver arrives knowing   SMS sent to Security Guard;   Stop skipped; route
customer is ready.       Driver drops box and leaves.  instantly re-optimized.

The Geofence Trigger: As the delivery van enters a specific radius (e.g., entering your housing society or neighborhood pin code), the driver's logistics app registers the GPS coordinate.
The Cloud Call: The app automatically commands the cloud server to call your registered mobile number. The driver does not see your number, ensuring data privacy.
The Interactive Menu: You pick up, and an automated voice says:

"Namaste, your Amazon/Flipkart delivery partner is arriving in 5 minutes with a package worth ₹1,200. Please keep your OTP ready. Press 1 if you are available, Press 2 to leave it with the gate/security, or Press 3 to reschedule to tomorrow."

Why IVR is Much Better for India Than Live Maps

For the unique infrastructure and cultural landscape of India, IVR vastly outperforms a live GPS map:

Bypasses Poor Internet Context: Live map tracking requires both the driver and the customer to have a rock-solid, high-speed 5G/4G data connection. In deep basements, elevator shafts, or patchy network areas, maps freeze. IVR operates over standard cellular voice networks, which work almost everywhere.
The "Gatekeeper" Filter: In modern Indian apartments, delivery agents are often stopped at the main gate by security guards using apps like MyGate or NoBrokerHood. An IVR call gives the customer a 5-minute head start to "approve" the entry on their society app, preventing the driver from getting stuck at the gate.
Zero Cognitive Load for Drivers: Last-mile drivers in India navigate chaotic traffic, dodging two-wheelers and potholes. They cannot constantly check a screen to see if a customer is messaging them. IVR collects the customer's response (e.g., "Leave with security") and updates the driver's screen as a simple, actionable instruction.
Solves the OTP/COD Bottleneck: If a delivery requires a secure cash payment or a mandatory 4-digit OTP, the IVR call acts as a psychological alarm clock. It prompts the customer to find their phone, look up the SMS, or grab exact cash before the driver knocks. This drops the time-per-stop from five minutes to under 60 seconds.

The Financial Advantage

For logistics giants like Delhivery or Xpressbees, making a 5-second automated cloud call costs a fraction of a paisa. Conversely, maintaining a data pipeline that continuously streams millions of real-time GPS locations to user apps costs massive amounts in cloud server infrastructure.

By substituting visual tracking with voice automation, Indian carriers have built a highly cost-efficient system tailored perfectly to the local ecosystem.

An automatic proximity alert (like a message when the package is 500 metres away) (highlighted in your PDF) is actually one of the most practical and elegant solutions for the Indian market. It solves the customer's anxiety without revealing the driver's exact movements or overwhelming them with calls.

In fact, the technology to do this exists, but implementing it purely by physical distance (like "500 metres") causes issues in Indian cities due to high vertical density.

Why a Strict "500-Metre" Radius Fails in India

The Apartment Complex Problem: In dense urban areas like Mumbai or Bengaluru, a delivery agent could be at a high-rise society right next to yours. The GPS says they are only 100 metres away, but they have 15 packages to deliver in that society first. If you get a "500m away" alert, you might wait at the door for an hour while they walk between towers.
The Crow-Flies vs. Road Distance: In tightly packed residential layouts or across divided main highways, a driver might be 200 metres away "as the crow flies," but because of U-turns or one-way streets, it could take them 15 minutes to actually drive to your gate.

The Real Solution: A Hybrid "Smart Notification" System

To make this work seamlessly in India, logistics companies can deploy a system based on stops and time, rather than pure physical distance. A highly functional blueprint includes:

1. The "Next Stop" Triggered SMS / WhatsApp

Instead of a continuous GPS fence, the system links directly to the delivery agent's application.

How it works: When the agent scans and marks package #44 as "Delivered" at the previous house, the system instantly triggers an automated WhatsApp message or SMS to you (package #45): "Your agent has just completed their last stop and is heading to you next."
Why it works: This gives you a highly accurate 3-to-7 minute heads-up without exposing a live map.

2. Interactive Interactive Voice Response (IVR) or Automated Flash Calls

Some Indian logistics companies and quick-commerce apps use automated bots right before arrival.

How it works: When the driver enters a geofenced area near your neighborhood, a system-generated call rings your phone: "Your delivery agent is arriving in 5 minutes. Please keep your Cash/OTP ready."
Why it works: It bypasses internet connectivity issues and immediately alerts the customer to grab their phone or head to the gate.

Part B · Gnani quick start (from `quick-start.md`)

STT REST: POST https://api.vachana.ai/stt/v3, formats WAV, MP3, OGG, FLAC, AAC, M4A, 8 kHz to 44.1 kHz, max 60 seconds, language_code=hi-IN, model gnani-prisma-v2.5.
STT Realtime: wss://api.vachana.ai/stt/v3/stream, binary PCM frames (16-bit mono, 8 or 16 kHz, 1024 bytes per frame), transcript events with latency.
TTS REST: POST https://api.vachana.ai/api/v1/tts/inference, model timbre-v2.5 (recommended), voice e.g. "Nalini", language hi-IN.
TTS Streaming: POST https://api.vachana.ai/api/v1/tts/sse (Server-Sent Events, base64 audio chunks).
TTS Realtime: wss://api.vachana.ai/api/v1/tts.
Voice cloning: step 1 POST /api/v1/tts/voice-clone/embeddings (reference audio, ideally 5 to 30 seconds); step 2 synthesize with model vachana-vc-v1 and the speaker_embedding.
Batch transcription: Batch STT Introduction (linked in the doc).

Part C · Research notes from earlier in our chat (mine, to verify)

warning · - Gnani: quick start shows STT, TTS and voice cloning only. No telephony, no documented sentiment signal in what we read, Hinglish accuracy unconfirmed. Telephony needs a partner such as Exotel, Ozonetel or Knowlarity. - Delhivery: public API list covers Pincode Serviceability, Fetch Waybill, Warehouse create/update, Shipment create/update, Tracking, Shipping Cost, Label, Pickup Request, NDR Package Action. GPS appears at status changes only; no geofence primitive known; the developer portal needs a Delhivery One login. - Pine Labs: individual-KYC built; no household delegation, provisional credit or trust scoring known. - Fourth rail guess: identity / household registry (Setu was a guess).

Back to 00 README Index