SAGA Development

SAGA Development’s sales office answered the same questions about its Rybalsky complex every day: ceiling height, parking, floor plans, payment terms. Buyers expected instant replies in Messenger, including outside office hours, and they typed questions in their own words rather than tapping buttons. The developer wanted an assistant that understands free-text Ukrainian and Russian, presents the complex, qualifies the buyer and hands warm leads to managers with full context.

The bot runs on the in-house BotMother PHP framework and greets visitors as the complex itself. An image-card carousel covers news, construction progress, technology, plans and purchase terms, and a six-step questionnaire qualifies the buyer. Typed questions go to a Python NLP service (Flask, gensim embeddings, pymorphy2 lemmatisation, spell-checking) matched against an admin-managed Q&A base with per-question thresholds; uncertain ones are logged for review. A daemon pushes leads to Bitrix24. I designed the NLP service and the CRM integration.

Free-text answers via NLP with per-question confidence thresholds

Floor plans by room count and a six-step apartment questionnaire

Bitrix24 lead sync daemon and Facebook Custom Audiences tagging

Handover to a live manager in the Page inbox and back to the bot

Laravel + Voyager admin for texts, Q&A, plans and tracked links

Bot, admin and NLP service delivered over two years; the core was later reused for KAN Development

PHP (BotMother)

Messenger Platform

Python

Flask

gensim

pymorphy2

Bitrix24 API

Laravel / Voyager

MySQL