How Mandali uses AI

The useful part is not the model. It is what the community produces: verified work histories, real service outcomes, and what stores actually buy. AI is how Mandali will turn that into something an owner can use at 11pm, in the language they speak.

The rules we build by

• People decide, AI drafts. Mandali will not screen out, rank away or reject a candidate for an owner. It will shortlist and explain, and the owner chooses. The same holds for suppliers and service providers.
• Your store's numbers stay yours. Anything Mandali shows the community will be aggregated and anonymous. No member will see a named store's numbers.
• In the language owners speak. Gujarati, Hindi and English, by voice or by text, so nobody has to write in a second language to get help.
• Nothing pretends to be a person. AI answers are labelled as AI, and a human is always one step away.
• Answers show their source. If AI says a provider is reliable or a price is high, it points to what that came from.

Where AI fits

1. 1

   Hiring, on both sides

   An owner will describe the opening in their own words and get a shortlist with reasons attached, such as two years at a store nearby, verified by another member. A candidate will speak their work history instead of writing a resume, and see the jobs that match their shift, distance and language. The owner still makes the hire.

2. 2

   Support that answers at 2am

   A text support assistant in the owner's language, answering the common questions instantly and handing over to a person for anything involving money, an account, a dispute or someone's job. The person receives the full conversation, so the owner never repeats themselves.

3. 3

   Services that match the job

   An owner will describe the problem, such as a walk-in cooler that stopped overnight. Mandali will turn that into a proper request, send it to providers who cover that area and that trade, and lay the replies side by side on scope, response time, price and what other owners said about them.

4. 4

   Procurement that sees it coming

   Once ordering runs through Mandali, AI will predict the next order, point out products that are selling well in similar stores nearby, and show where margin is leaking. This one depends on order history, so it follows the ordering layer rather than leading it.

5. 5

   Community and news

   Translation across all three languages so a post written in Gujarati is useful to an owner who reads only English, short summaries of long regulatory updates, and scam screening on marketplace listings.

What we will not do

• No automated rejection. An owner reviews every candidate. That is also how the AI hiring rules in New York City, Colorado and Illinois expect these tools to be used.
• No AI posts pretending to be an owner in the community feed.
• No selling store-level data, to anyone.
• No AI-written reviews of service providers. Ratings come from owners who used them.

AI is a tool here, not the product. The product is the community. See what we build and in what order.