Platform · AI agents · WhatsApp commerce

In development

A team of AI staff that answers, books, sells and chases payment on WhatsApp, with people in charge.

A multi-tenant platform for small and medium businesses whose customers reach them on WhatsApp. AI agents answer every chat from the business's own knowledge, book appointments, score leads and follow up on invoices, and hand over to a person when one is needed. Behind the inbox sits the system that runs the business: CRM, calendar and rosters, jobs and quotations, stock, invoicing and automation.

Client
Northridge Technologies
Country
Singapore
Services
Applied AI and LLM systemsCustom business systemsSystems integrationCloud architecture and DevOps
allset / inbox
Interface illustration with sample data. No client records are shown.
8
AI staff roles, from front desk to owner's assistant
589
automated backend tests
5
integrations: Google, Shopify, TikTok, Lalamove, Do Not Call registry
1
gateway every AI call passes through, metered for cost
01

AI staff

Eight roles that answer, book, sell, market and collect, each with its own workload.

02

Linked chats

Buyer and seller side by side. The AI suggests what to pass on; staff press Forward.

03

Back office

CRM, calendar, jobs, stock and invoicing that the agents can act on.

04

Automation

Flows with triggers and safety limits, plus analytics on what they earned.

The collaboration

AllSet is developed in collaboration with Northridge Technologies of Singapore. It is aimed at service businesses that live on WhatsApp: aircon servicing, property agencies, clinics and salons, tuition centres, workshops. For them the chat is the shopfront, and every message answered late is a sale lost.

AI staff, by job

The AI is presented the way an owner thinks about a team: by role. Each role has a job, a status, and a count of what it did this week and what needs a person. The owner sees at a glance that seven of eight roles are working and nineteen things need attention.

  • Front desk answers every chat from the business's knowledge, day and night, and hands over when a person is needed
  • Bookings finds free times, books, moves and cancels appointments in the chat
  • Sales reads every chat for lead details, scores leads and makes sure hot ones get a fast answer
  • Marketing welcomes people from QR codes, ads and posts, and wins back customers who went quiet
  • Accounts tells customers what they owe, sends the payment QR, reads payment screenshots and chases unpaid bills
  • Operations, a quality coach and an owner's assistant over WhatsApp complete the team

How the agents are built

This is agent engineering, not a prompt in front of a chat box. Agents call tools to look things up and take actions: check a calendar, create a booking, raise a quotation. They answer from an indexed knowledge base using retrieval over embeddings, and report gaps when they cannot find an answer. Every model call in the product goes through one gateway, which records what it cost.

  • Tool-calling agents with a defined set of skills, each gated by what the business has switched on
  • Retrieval over the business's own sources with pgvector, keyword search as a fallback, and a list of knowledge gaps
  • The model chooses from lists the code gives it; code never tries to match free text
  • Agent behaviour is versioned: a draft until published, with the published version serving customers
  • Mystery-shopper test runs and staff corrections to check and improve an agent before and after it goes live
  • Handover rules: when staff take over a chat the AI pauses, and it can be told to continue
  • Customer photos and PDFs are read by the model, such as a payment screenshot sent in the chat

The inbox, with the lead beside it

Staff work in a shared inbox. Beside each conversation sits what the system has learned: a lead score with the reasons behind it, the stage, what the customer wants in their own words, and qualification details the AI extracted with its confidence shown. WhatsApp's own rules are built in, so when the reply window has closed the composer says so and offers an approved template.

The go-between: linked chats

Some businesses sit between two parties: a property agency between seller and buyer, a tuition centre between parent and tutor. AllSet links the two WhatsApp conversations so staff see them side by side. When the seller writes something that matters to the buyer, the AI flags it as meant for the other side and drafts a clean version to pass on. The agent presses Forward, or dismisses it. The two parties are never connected to each other, and nothing is sent without a person pressing Send.

  • Two or more chats linked around a deal, each labelled with its role, such as Buyer and Seller
  • The AI spots a message meant for the other party and rewrites it in a neutral, professional tone
  • Staff always decide: forward the suggestion, edit it, or dismiss it
  • A simple log of the amounts offered, kept alongside the conversation
  • The parties never see each other's number or messages directly

The business behind the chat

An agent that can only talk is a novelty. AllSet's agents act on a real back office, which is most of the codebase.

  • CRM with scoring, routing, pipelines and deals that move as quotes and bookings happen
  • Calendar and rosters by person, room and vehicle
  • Quotations, job orders, visits, parts and billing
  • Catalogue, stock and purchasing, counter sales, invoicing, packages and recurring billing
  • An automation engine with triggers, steps and safety limits, and a way to describe a flow in words
  • Analytics on what chats, AI and flows brought in: money collected, bookings, reply speed, funnel
  • A public API, webhooks and connectors for Google, Shopify, TikTok, Lalamove and the Do Not Call registry

Engineering

The backend is layered, and the layering is enforced by an import linter in CI. Every push runs linting, type checks, unit tests, the build and a code-quality scan before deploying to the development environment. There are 589 automated backend tests, including agent tests that run against a scripted model so they never call a real one.

Where it stands

AllSet is in development and running on a development environment with demonstration businesses. The screenshots on this page are from that environment and show fictional companies and customers.

  • React 19
  • TypeScript
  • TanStack
  • Tailwind CSS
  • Django 6
  • PostgreSQL + pgvector
  • Claude
  • WhatsApp Business API
  • AWS
  • Terraform

Inside AllSet

The product so far.

Screenshots from the development environment, showing demonstration data.

AllSet: dashboard
dashboard
AllSet: linked chats buyer and seller
linked chats buyer and seller
AllSet: shared inbox with lead scoring
shared inbox with lead scoring
AllSet: ai staff
ai staff
AllSet: deals pipeline
deals pipeline
AllSet: analytics
analytics

More work

All work
clausework / mortgage-debtors
Legal · Banking litigation

Clausework

From a bank's spreadsheet to court-ready documents, with every step on the record.

Kek Ooi & Lee Hong
iqc / quality-control
Manufacturing · Water reticulation

IQC

Every product traced from the factory floor to the purchaser, with a certificate anyone can verify.

Soon Chuan Hong Hardware Sdn. Bhd.
question-bank / review-jobs
Education · Assessment content

Question Bank AI

Textbooks in, reviewed exam questions and learning cards out, in English and Bahasa Melayu.

Tavis

Start a conversation

Facing something similar?

Tell us what is slowing the business down. You will talk to an engineer, not a salesperson, and get a straight answer on whether software is the right fix.