I have spent years close to how work actually gets done here. First on a finance desk. Then inside a company that delivers groceries in ten minutes. Then on shop floors, where the whole business lives in one WhatsApp group.
Here is what I learned. The work is real and the people are sharp. The software rarely fits. It asks them to change. They do not, and they are right.
So I build AI that fits the way the work already happens, at ten orders a day or ten lakh. It speaks the language. It handles the mess. It asks before it touches money.
I own the customer experience after the order is placed. When an item is missing, a refund is due or a delivery goes wrong, the systems that respond are the ones I build.
Dispute agents
A customer with a missing or wrong item should not wait in a queue for a person to look at a photo.
Agents that read the complaint and the photo, check the order and settle it in the same chat. When a case is unclear, they hand it to a person.
Fraud
Some refund claims reuse the same photo across orders and accounts.
A check that compares each dispute photo with recent ones and flags a reused image before the refund goes out.
QA surveillance
Support quality was judged on a small manual sample. Most conversations were never reviewed.
AI that scores conversations at full volume against the support rules, weighted by how critical each rule is, in Hindi, English and regional languages.
Voice of customer
Thousands of complaints a day say the same thing in different words, and no single team owns the fix.
An LLM pipeline that reads every wrong-item complaint, groups it by cause and sends each pattern to the team that can fix it. The daily report is read up to leadership.
WhatsApp ordering
For many customers, sending a message is easier than searching an app.
An agent that takes an order in plain chat, in the customer’s own words, and turns it into a cart they confirm.