A conversational sales agent that chats with leads in English, Hindi, or Hinglish, qualifies them, handles objections, books site visits, hands off to a human when needed, and gives the business a scored, structured summary of every conversation.

Demo: a real-estate agent ("Priya" from the fictional Northstar Homes, Sector 79, Gurugram). The same design can be adapted to clinics, coaching institutes, e-commerce, or any business that gets the same enquiries all day.

Leads message at all hours, and slow or inconsistent replies lose sales. Many Indian customers prefer Hindi or Hinglish, which most chatbots handle badly. And even when a chat goes well, the sales team still has to read the whole thing to know whether the lead is worth a call.

Speaks the customer's language. Detects and mirrors English, Hindi, or Hinglish, even if the customer switches mid-conversation.

Qualifies naturally. Works budget, configuration, timeline, and purpose (own use or investment) into the conversation instead of running a checklist.

Sticks to the facts. Never invents prices, amenities, floor plans, or possession dates, and says so when it doesn't know.

Handles real situations. Price objections, busy customers, callback requests, and "stop contacting me" are all handled correctly.

Books site visits. A booking tool validates 10-digit mobile numbers and visiting hours (9 AM to 6 PM), returns a booking reference, and recovers or escalates to a human if a slot fails.

Hands off to a human for complex or sensitive conversations.

Streams replies live, word by word, like a real chat.

Keeps multiple chats. Switch between past conversations from the sidebar.

Lead score from 1 to 10 based on a clear rubric (booking completed scores highest, objections lowest).

Structured summary: customer details, budget range, preferred configuration (2 BHK / 3 BHK), key objections, language preference, follow-up flag, and a short executive summary, returned as JSON for a CRM or dashboard.

8 automated end-to-end test conversations covering the happy path, price objection, busy customer, stop request, unknown question, Hindi conversation, human escalation, and invalid booking details. They run from the UI or through an API endpoint.

Real estate lead qualification

Clinic and coaching-institute enquiry handling

E-commerce pre-sales support

First-response bots for small businesses that get repeated enquiries

Python, LangChain (tool-calling), FastAPI (Server-Sent Events streaming), Streamlit, Gemini / Groq LLMs

The booking system is simulated (including random failures to test error handling), and conversations are stored in memory for the demo. In a real project, bookings would connect to your CRM or calendar, and sessions would be saved in a database.

I build custom chatbots and AI agents for small businesses, in English, Hindi, and Hinglish. Contact: rishibarapatre@gmail.com