‹ All deployments
Casagrand · Real estate · 2026
A sales agent that knows every flat in Chennai
A buyer calls and asks for a 3BHK near OMR, ready by next year, under a crore. The agent has to know what's actually available today, not what was in last month's brochure. Then it qualifies the buyer and books the site visit.
Team servedSales
ChannelInbound and outbound voice
Coverage63 projects in Chennai
My roleCustomer success, data pipeline

The problem
The agent is only as good as its inventory, and the inventory lives in the sales team's ready reckoner. It's a spreadsheet that changes every week, with merged cells, sold-out notes typed inline, prices in different columns, and the same locality spelled five different ways. Feed that to an agent raw and it will pitch a flat that sold last Tuesday.
What I did
- Wrote the pipeline that turns the ready reckoner into one clean record per unit type, then groups it by locality, project and unit, the way a buyer actually asks.
- Built locality matching for how people talk. Misspellings map to one canonical area, and umbrella terms widen the search: say "Tambaram" and the agent also looks at East Tambaram.
- Flagged every sold-out unit so the agent never pitches one, but can still say the project exists.
- Made each run produce a data-issues report for the sales team: missing prices, duplicate rows, rates in the wrong column. The source sheet got cleaner every week.
- Designed the conversation. Qualify on location, BHK, budget and timeline, then book the site visit, with a dashboard for the sales team tracking intent and visits.
How it works
Ready reckoner›Clean + group›Knowledge base›Voice agent›Qualify›Site visit booked
Impact
63
projects the agent can talk about
193
unit types, each with live status
30
localities, matched the way people say them
Award
Voice AI Pioneers
Stack