Conversational AI that speaks your customer's language
98%
Reduction in manual call handling for routine inquiries
12x
Increase in concurrent support capacity without hiring
Sub-5s
Average end-to-end voice response time per customer turn
Call volume was scaling faster than the team
Zero Lifestyle, Pakistan's fastest-growing tech accessories brand with over 1,000,000 customers, was facing an unsustainable surge in inbound support calls. Customers calling to check order status, ask about product features, or inquire about delivery timelines were overwhelming a small human support team. Each call required the agent to manually search a fragmented knowledge base of product specs, shipping policies, and return rules — leading to long hold times, inconsistent answers, and missed upsell opportunities. With a catalog of 20+ smartwatch, earbud, and headphone SKUs — each with unique pricing, features, and stock status — even well-trained agents struggled to keep response times under two minutes.
A real-time AI voice pipeline purpose-built for Pakistani support
Neura Agency architected a fully serverless voice agent using LiveKit's real-time infrastructure. The pipeline chains together Groq Whisper (speech-to-text with Urdu support), Google Gemini 2.0 Flash with OpenRouter fallback (LLM reasoning and tool calling), and custom Microsoft Edge TTS (text-to-speech with the ur-PK-UzmaNeural voice). A Silero VAD model handles natural turn-taking — no push-to-talk required. The agent was equipped with a suite of function tools: product search and detail lookup across 20+ SKUs, order status queries, warranty and returns policy retrieval, and FAQ resolution. The entire stack — from browser WebRTC frontend (Next.js 15) to SIP telephony bridge (FreeSWITCH) — was designed for sub-5-second response times and 99.9% call reliability
Core Capabilities
Real-Time Urdu Voice Conversation
End-to-end voice pipeline with Groq Whisper STT, Gemini LLM reasoning, and Microsoft Edge Urdu neural TTS. Natural turn-taking via Silero VAD with sub-second latency.
Product Catalog & Intelligent Search
Customers can ask about any product by name, category, or feature. The agent searches a 20+ SKU catalog, returns pricing with discount percentages, key specs, review counts, and stock status — all in natural Urdu speech.
Order Tracking & Status Lookup
Instant order status retrieval by order ID or courier tracking number. The agent validates the ID, fetches the current location, courier, warehouse status, and estimated delivery date without human intervention.
Multi-Channel Voice Support
Dual connectivity path: browser-based WebRTC calls via the Zero Lifestyle website and traditional PSTN phone calls routed through FreeSWITCH and LiveKit SIP — providing universal customer access.
Intelligent Call Management
Context-aware human agent transfer, automatic end-of-call detection, and comprehensive per-session metrics collection (STT latency, LLM tokens, TTS duration) for continuous performance monitoring.
Policy & FAQ Automation
Dynamic retrieval of warranty terms, replacement policy, shipping timelines, payment methods, and frequently asked questions — eliminating manual lookup and ensuring consistent, accurate answers every call.
Architecture & Technology
Frontend
- Next.js 15
- React 19
- TypeScript
- LiveKit Client SDK
- CSS Custom Properties
Backend
- Node.js
- LiveKit Agents SDK
- LiveKit Server SDK
- Microsoft Edge TTS (msedge-tts)
- Silero VAD
Infrastructure
- LiveKit Cloud
- Docker
- FreeSWITCH (SIP/PSTN bridge)
External APIs
- Groq API (Whisper large-v3-turbo — STT)
- Google Gemini 2.0 Flash / Gemini 2.5 Flash (LLM)
- OpenRouter API (LLM fallback provider)
- Zadarma or equivalent SIP provider