Building an AI-Native Voice Platform for Indian Businesses

From Voice Calls to Business Actions

Indexnine worked with Voca Labs through its Startup Lab approach, combining product thinking, architecture and AI engineering to shape the MVP around a shared core platform.

Rather than building separate products for every industry, the architecture was designed around a common voice and intelligence layer with configurable vertical packs.

• AI agent scripts
• Structured information extracted from calls
• Workflow templates
• WhatsApp and communication actions
• Industry-specific dashboards

This architecture allowed Voca Labs to target multiple use cases without creating a separate codebase for every vertical.

Engineering AI Into the Product

The platform was designed around several interconnected capabilities:

• Voice AI handled speech-to-text, text-to-speech and conversational interactions across Indian languages and Hinglish.
• Call Intelligence transformed conversations into structured business information—including intent, dates, amounts, sentiment and disposition.
• Workflow Automation converted those insights into actions such as WhatsApp messages, CRM updates, tasks, callbacks and follow-up calls.
• Compliance was embedded directly into the calling flow, with requirements around DND checks, consent, calling windows, disclosures and audit trails.

Claude-Powered Intelligence

As part of the AI architecture, Anthropic’s Claude API was used to support the intelligence layer, helping transform unstructured conversational information into useful, structured business context.

Claude’s capabilities were particularly relevant to the product’s requirement for extracting meaning from conversations rather than simply transcribing them—supporting the broader intelligence layer around summaries, intent, structured fields and downstream actions.

Claude Research was also leveraged as part of the research and data-intelligence workflow, helping accelerate exploration and synthesis of information required to shape the product and its AI-driven capabilities.

This created a broader AI architecture in which voice interaction, reasoning, structured extraction and workflow automation could work together rather than functioning as isolated AI features.

Designing for India’s Voice-First Reality

A major product requirement was making AI calling feel natural for Indian users.

The MVP specification included Hindi, English and Hinglish by default, with support for additional Indian languages, language switching during calls and mixed-language conversations such as Hinglish and Marathi-English interactions.

The platform was also designed to move beyond transcription. Each call could produce a summary, structured fields and an automatic disposition, allowing the conversation to become an actionable business event.

Three Industry Use Cases. One Intelligent Core.

Real Estate

AI agents qualify property leads, answer questions, capture requirements and help schedule site visits.

Insurance

AI agents support renewal and enquiry workflows while handing sales conversations to licensed human agents.

Invoice Collection

AI agents manage payment reminders, capture promises to pay and trigger follow-ups based on payment commitments.

The workflow engine then connects these conversations to actions such as WhatsApp communication, CRM updates, task creation and future AI calls.

Built With Scale and Compliance in Mind

The MVP was designed with production considerations from the beginning rather than treating them as post-launch additions.

• Encrypted data storage and India-region data handling
• Role-based access and auditability
• Consent management and calling-window enforcement
• AI disclosure requirements
• Compliance-aware calling workflows

The architecture was also designed to remain flexible across AI providers, with a Pipecat-based pipeline allowing STT, TTS and LLM providers to be swapped as the platform evolved.