Startup Buildathon India 2026 · Submission Dossier

A phone call can be an AI interface.

Ph2AI brings an inbound voice assistant and a translated-call flow to ordinary telephony.

01 / Problem statement

AI access still assumes a screen, an app, and a data connection. The phone network reaches people; the AI interface often does not.

950M

people in the 2022 mobile internet usage gap used only basic services.

600M

of that group had a basic or feature phone.

Source: GSMA, State of Mobile Internet Connectivity 2023. Context figures, not Ph2AI market size.

02 / Product idea

Two call paths. One familiar interface.

01 / VOICE ASSIST

Ask a question out loud.

A caller reaches the phone AI assistant and hears a spoken response. The assistant test recording is included in the demo.

02 / TRANSLATED CALL

Talk across languages.

Ph2AI connects a second call leg and routes the two speech directions through the translation flow. The founder validated an English–French call.

03 / How the call works

Phone line to voice, and back.

Caller dials the configured Exotel number.
Exotel streams call audio to the service.
Cloud Run manages the voice or translation session.
Spoken audio returns over the phone call.
TRANSLATIONTwo telephone legs carry the English caller and French recipient. The latest relevant Exotel record shows both legs complete for 181 seconds.
TECH STACKExotel · Google Cloud Run · Gemini sessions · Firestore preferences

04 / Prototype validation

Both flows were tested on calls.

181SECONDS CONNECTED
VOICE ASSISTThe supplied recording captures an inbound conversation with the phone AI assistant.
TRANSLATED CALLThe founder confirms an English-speaking caller and a French-speaking recipient spoke through the bridge.
EXOTEL RECORDBoth legs completed; 181 seconds of connected conversation.

05 / Market context and potential impact

Extend spoken AI through the network people already use.

Potential users include multilingual families, local businesses, and field teams that rely on voice calls.

6.45Cr

MSME registrations reported by India’s Ministry of MSME as of 11 June 2025. This indicates a potential ecosystem for pilots, not Ph2AI customer demand or TAM.

Source: Ministry of MSME RAMP portal, including Udyam Assist registrations.

06 / Cost snapshot

₹3.68 per connected minute in the current prototype.

₹578.67

Nov 20, 2026: ₹522.35 Google Cloud actual cost before credits + ₹56.32 Exotel API-reported price, divided by 157.1 connected minutes. This early build uses extra compute to keep latency low; ₹3.68 is a prototype baseline, not a target price. Next: optimize compute, use deterministic services where feasible, scale, and explore lower-cost 1800 access. Government campaigns/services and SME translation can be sponsor-funded; users may pay for on-demand AI assistance. Pilot data and expert guidance will shape the viable model.

07 / Next development

Turn one successful test into a repeatable pilot.

Repeat both flows with more callers.
Ask each person what they understood.
Measure call quality and consent.
Calculate phone and cloud cost together.