How Marriage.pk actually works

The platform, explained honestly.

Plain-English documentation of every agent, pipeline, and moving part — written from the code and verified against the running system, never from memory. If a page says something works, it was tested the day the page was stamped.

flowchart LR
    CUST["👤 Customer
WhatsApp"] <--> GW["📨 whatsappGateway
every message, one door"]
    GW <--> MAI["🤖 marriageAI
replies to every message"]
    OAI["📣 outreachAI
reaches out on its own"] --> GW
    MM["💞 matchmakingAI"] --> MD["proposal delivery"] --> GW
    WEB["🌐 marriage.pk website"] --> API["⚙️ coreAPI
api.marriage.pk"]
    DASH["🖥️ Staff Dashboard"] --> API
    DASH --> INBOX["🙋 Agent Inbox
human takeover"]
    MAI -.-> INBOX
    LEARN["📡 learning pipelines"] -.-> MAI
    classDef hub fill:#e8eef4,stroke:#4e5d6b,color:#182420
    classDef agent fill:#e7f2ec,stroke:#166b4e,color:#182420
    class GW,API hub
    class MAI,OAI,MM agent

The whole system on one map. Every arrow is a real, verified path — nothing aspirational.

Browse by area

The conversation agent

marriageAI

The AI that answers every customer on WhatsApp — how a single message becomes a reply, which brain handles it, and what it knows about you.

2 pages · includes the Question Bank pipeline

The proactive agent

outreachAI

The AI that decides — like a human would — whether, when, and what to message customers who haven't written to us.

1 page

Getting smarter

Learning pipelines

The six feeds that teach the agents from real conversations — which ones genuinely work, and what each one does.

1 page · live status board

The front door

whatsappGateway

Every WhatsApp message, in and out, passes one door — recording, safety checks, and the free testing band that can never touch a real customer.

1 page

🧭

How to read these docs. Every page carries a "verified" date. Diagrams show what the code does today, not the original design. Where something is switched off or waiting on a decision, the page says so plainly.