Our product · Real estate software · India · 2026
LeadOne AI
A sales CRM for Indian builders that we co-founded and build ourselves: an iPhone and Android app for the reps on site, LeadOne AI HQ on the web for the office, and the marketing site at leadoneai.com.
- What we did
- Product and design, Mobile app, Web app, Backend, Marketing site, App store releases
- Stack
- Flutter, Next.js, React, TypeScript, Supabase, Postgres, Meta Lead Ads, Razorpay
- Live
- leadoneai.com
The brief
LeadOne AI is our own product, not client work. Vedant co-founded it with Manas under One Construction, a company in Nashik, and builds it: the app, the web workspace, the backend and the site. Everything on this page is in the product today or on leadoneai.com.
It started from how sales runs at most builders. Leads from Facebook and Instagram ads sit in a Google Sheet until someone gets round to calling. Which flats are still available lives on a whiteboard at the site office or in a WhatsApp group. Reps work out cost sheets by hand, and a discount turns into a phone call to the owner. Once a flat is booked, the money still owed is tracked in yet another spreadsheet, and demand letters are typed one by one.
So we built one system for the whole path, from the first enquiry to the last instalment, with two front doors: a phone app for the people who sell on site, and HQ, a desktop workspace for owners, sales heads and accounts.
One backend behind both apps
The app and HQ run on the same Supabase project: the same rows, the same server functions, the same access rules. A stage changed on a phone shows up in HQ, and a price changed in HQ shows up on the phone. Access is enforced by Postgres row-level security, not by the screens, so a sales rep never receives booked value, collections or quotes from the server, whichever app they open.
Business rules live in one shared package that HQ builds on: rupee formatting, quote and cost-sheet maths, payment milestones, which stage changes are allowed, and how towers turn into units. HQ is not allowed to re-implement any of them.
Around the database sit edge functions for the work that has to happen without anyone pressing a button. One receives Meta Lead Ads webhooks, checks their signature and files each lead under the right builder. Another runs every minute and sends "call this lead" reminders as push notifications.
An app for people who sell on site
Reps are on their feet, at the site office or in a sample flat, often with poor signal. The app is built for that: big touch targets, changes queued on the phone and sent when the connection comes back, and voice entry. A rep taps the mic and speaks a new lead or a task in English, Hindi or Marathi, and the form fills itself: name, phone, budget, next step. The day, time and action are read from the sentence on the phone, in Latin script, Devanagari or a mix; AI only fills what the parser could not.
The unit grid is the same live stacking plan the office sees: tower by tower, floor by floor, available, booked and sold. From a unit a rep can generate a quote, hold it for a lead or book it with a token.
HQ: the office side
HQ is where the owner and the sales head work. The stacking plan shows every unit of every tower, coloured available, booked and sold. Projects are set up here: towers and unit types, rates and charges, and payment milestones that have to add up to 100%.
Discounts have a rule instead of a phone call. Each project has a limit. When a rep’s quote goes past it, the quote is held with a "Request approval" button; an owner, admin or manager approves all of it, part of it, or rejects it with a reason, and only then can the quote go out.
Every change is written to an audit log, so an owner can see who moved a lead from Contacted to Site Visit Done, or who changed a price, with the before and after.
Collections that follow construction
In Indian construction-linked plans, buyers pay as the building rises. So in HQ, the accounts team marks a tower milestone done, say the fifth slab, and LeadOne AI lists who now owes what. Demand letters come out as PDFs, one at a time or as a ZIP, at the project’s GST rate, and each is recorded in the ledger as an upcoming payment. Reminders can go by WhatsApp.
The ledger shows agreement value, demanded, collected, outstanding and overdue, with ageing in 30-day buckets counted in India days. It exports as a CSV and as a Tally receipt file, so the accountant does not retype anything.
A site that shows the product working
leadoneai.com opens on a point-cloud model of a two-tower project that tells the whole story in 30 seconds: excavation, structure, sales, the stacking plan, collections, handover. On wide screens the hero stays pinned and scrolling scrubs the story; visitors can pick a phase or drag the model to turn it. Phones and visitors who prefer reduced motion get a still image instead, and the scene only loads once the page has painted and the browser is idle.
Further down, the HQ screens are recreated in the page and driven by a small script engine: a cursor moves to named elements, clicks and types, on a loop. The controls are real, so a visitor can take over, and the script waits for them.
Around that sit pages written to be found and read: features, comparisons with other CRMs, guides, a glossary, city pages, and pages in Hindi and Marathi.
Craft notes
Call recordings matched to the right lead
Android does not let apps record calls, so the app reads the files the phone’s own recorder saves and matches each one to a lead: by the last 10 digits of a number in the file name, then by a contact name only one lead has, then by the time of a call placed from the app. A file that matches no lead is never uploaded.
A hero drawn as a function of time
The point cloud is drawn on a plain 2D canvas, with no 3D library. Every frame is a pure function of the timeline position, so the same code plays, loops, seeks and scrubs on the home page, and renders the site’s film frame by frame.
Demos that never fake the layout
The site’s demo engine measures its target elements when each step runs, so the cursor lands on the real button at any size. Demos pause off screen, in hidden tabs and while the visitor points at them, and with reduced motion they show the end state with a Play button.
HQ next to its database
Each HQ page reads the database several times on the server, so its functions run in the same region as the database, and the reads on every navigation are batched together rather than awaited one by one.
Tested on both sides
The Flutter app and HQ each carry their own suites of unit and widget tests, and every HQ pull request runs lint, type checks, tests and a production build in CI.
Analytics that respect the visitor
The site loads analytics only after the page is interactive, and not at all for browsers that send Global Privacy Control or Do Not Track.
What’s next
A WhatsApp agent that answers new enquiries in English, Hindi or Marathi is rolling out, as the site says.
AI calling is in early access and not on sale; LeadOne AI does not call buyers today.
Quick answers
- Who built LeadOne AI?
- LeadOne AI is Kiviro Labs' own product. Kiviro Labs, a design and engineering studio in Nashik, India, designs and builds it.
- What did Kiviro Labs do on LeadOne AI?
- Product and design, Mobile app, Web app, Backend, Marketing site, App store releases.
- What is LeadOne AI built with?
- Flutter, Next.js, React, TypeScript, Supabase, Postgres, Meta Lead Ads, Razorpay.
- Where can I see LeadOne AI live?
- https://www.leadoneai.com.
