AI order desk · Built for Pakistan

Voice notes in.
Orders and khata out.

Shopkeepers order by WhatsApp voice note in Urdu and Punjabi. Awaz Order listens, builds the priced order, updates their khata and replies in Urdu — in seconds, not minutes.

Try the live demo See how it works
Urdu · Punjabi · English Open-source AI models Keeps working if AI is down
A distributor counter where a phone voice note connects to a packed grocery order AI processing
Listen Match Verify
Bismillah General Store
0:14

“Do carton Shan biryani masala, paanch tin Dalda… udhaar likh dena.”

Order #1042Added to khata
2 carton · Shan Biryani MasalaRs 19,000
5 tin · Dalda Ghee 5kgRs 14,750
TotalRs 33,750
آپ کا آرڈر کنفرم ہو گیا ہے۔ کل رقم: 33,750 روپے
Why distributors use it

Built around how Pakistani trade already works

No new app for shopkeepers to learn. They keep sending voice notes; your team stops typing them out.

Understands real voice notes

Urdu, Punjabi, Roman Urdu and English mixed together — including “dedh carton”, “darjan” and “nahi nahi, teen kar dein”.

Matches your catalogue

Every item is checked against your products and price list. The AI never invents a price; anything uncertain asks “Did you mean…?”.

Khata that updates itself

Credit orders and payments land in each shop's ledger automatically, so the balance due is always right.

Replies in Urdu

A WhatsApp-ready confirmation with items, total and balance — as text, or read aloud in an Urdu voice.

How it works

From voice note to confirmed order in four steps

The AI does the listening and typing. A person stays in control of every order.

Record or forward

Record in the browser, upload a WhatsApp voice note, or type the order.

دو کارٹن شان بریانی، پانچ ٹین ڈالڈا

AI agents understand it

A team of specialised agents listens, works out what the shopkeeper wants, extracts the order and double-checks it.

Review in one tap

Prices come from your catalogue. Uncertain items are highlighted with one-tap suggestions.

Confirm and reply

The order is saved, the khata updated, and an Urdu reply is ready to send — text or voice.

Multi-agent AI

Six agents, one job each

Every message passes through a pipeline of specialised agents. Each hands a structured result to the next, and every step is traced so you can see who decided what, and how long it took.

🎧

Listener

Whisper large-v3

Turns Urdu and Punjabi voice notes into text, primed with your brand names.

🧭

Router

gpt-oss-20b

Decides what the message is: a new order, a payment (“bees hazar bhej diye”) or a balance question.

📦

Extraction

Qwen 3.8 → gpt-oss-120b

Pulls out items, quantities, payment and shop, and runs in parallel with the Router to save time.

🗂️

Catalogue

RapidFuzz + price list

Checks every product against your catalogue and prices it from your list. Models never set prices.

✅

Verifier

gpt-oss-20b

A second model re-reads what was said and flags missed items, wrong quantities and ignored corrections.

💳

Credit

Khata ledger rules

Checks each shop's khata against its credit limit and warns before an order goes over it.

Architecture

Agents, a human in the loop, and no single point of failure

Each message flows through the agent pipeline below. If an AI model is busy, the next one takes over, and rule-based fallbacks keep orders moving even with no AI at all.

Awaz Order pipeline A voice note passes through six agents in order: Listener (Whisper), Router (gpt-oss-20b), Extraction (Qwen then gpt-oss-120b), Catalogue (RapidFuzz and the price list), Verifier (a second model) and Credit (khata limits), then the Reply and Voice agents send an Urdu confirmation. If the AI is busy, offline fallbacks feed the Catalogue agent. A person reviews and confirms every order. if AI is busy 🎧 ListenerWhisper v3voice → text 🧭 Routergpt-oss-20border · payment · balance 📦 ExtractionQwen → gpt-ossitems as JSON 🗂️ CatalogueRapidFuzzmatch + prices ✅ Verifiergpt-oss-20bsecond opinion 💳 Creditledger ruleskhata limits 💬 ReplyUrdu text+ ElevenLabs voice Offline fallbackskeyword router · offline parser 👤 Human reviewconfirm · record payment
Specialised agentsEach agent has one job and its own model: Whisper listens, gpt-oss-20b routes and verifies, Qwen and gpt-oss-120b extract.
AI understands, code decidesPrices, totals and balances always come from the database — the model can't invent a number.
Designed not to breakCached samples, a model fallback chain and an offline parser keep orders flowing when an API is busy.
Powered by open-source AI

The technology behind every order

Open models, open tools — no vendor lock-in.

W
Whisper large-v3Speech-to-text
Q
Qwen 3.8 · 27BOrder extraction
G
gpt-oss-120BFallback model
R
gpt-oss-20BRouter & verifier
RF
RapidFuzzCatalogue match
EL
ElevenLabs v3Urdu voice
FA
FastAPIBackend
GQ
Groq CloudAI inference
🚀 Coming Soon

RAG · Agentic Workflows · WhatsApp API · Voice Cloning

We're building the next generation: RAG-powered search, LangGraph agent orchestration, WhatsApp Business API integration, custom voice cloning with ElevenLabs, ML pipelines for demand forecasting, and more.

See the full roadmap →

Try it with your own voice

Sign in with the demo account demo@awazorder.pk / Demo@1234, record an order in Urdu or Punjabi, and watch it become a confirmed order.

Open the demo