This project began as an internal initiative at Devexperts to automate trader interactions via chatbots, combining automation with human handoff in messaging channels. Project highlights 16 B2Bclients tested the AI widget in production95,000weekly unique end users144,000monthly responses sent by Devexa23.6Mmessages sent through the AI widget Searching for product-market fit Without an anchor client at the start, we iterated while validating demand across several use cases: Voice assistants (Alexa, Google Assistant) Support and education bots Broadcasts, signals, and surveys User-behavior analytics on trading platforms Each cycle narrowed the scope. The strongest signal came from embedding smart functionality directly into trading platforms. 2021 version of human support in the admin platform 2023 version of human support in the admin platform 2025 version of human support in the admin platform Platform integration and first client The turning point was a shift from standalone assistants to an in-platform widget focused on engagement and retention. Our first commercial partner confirmed the direction. In parallel, we adapted the tool across the Devexperts ecosystem (DXcharts, DXtrade, dxFeed, and others). Example of a widget embedded in a client’s trading platform, illustrated with DXtrade Designed for white-label customization From the first widget versions, we prioritized white-labeling so brokerages could match brand guidelines quickly. Robust design tooling sped up handoffs and ensured consistent theming across deployments. Deep customization allows the AI widget to be fully reskinned to match any brand identity Deep customization allows the AI widget to be fully reskinned to match any brand identity Deep customization allows the AI widget to be fully reskinned to match any brand identity UI evolution: from minimalism to practicality As usage scaled, the interface evolved to support real-world operations: Moved the main menu from top to top-left to preserve vertical space for chats and lists. Replaced accordions with a flatter structure for clarity and better screen use. Switched from underline-only inputs to filled fields for compact forms and legibility. Enhanced feedback with labeled reactions (“Thank you,” “Useless”) alongside icons to encourage quick responses. Added keyboard navigation (Tab) across forms to improve accessibility. UI improvements UI improvements UI improvements UI improvements UI improvements UI improvements Development process and release strategy With a flexible environment and continuous deployment, we followed an iterative approach: launch minimal prototypes, collect feedback early, and expand based on real usage. Rigorous design reviews preserved quality while enabling rapid experimentation. Feed AI Assistant Feedback Announcements The result What began as a chatbot concept became a modular, AI-powered engagement engine embedded directly in trading platforms. It supports: trader retention and re-engagement, personalized content and support, integrated analytics and trading operations, and seamless white-label deployment for brokers. Functionality spans both the admin application and the user-facing widget. Conversations Skills Knowledge Base Broadcasting Polls Events Feedback Log Trends in Analytics domain Symbols in Analytics domain Business Metrics in Analytics domain Users in Analytics domain Symbols in Analytics domain Trends in Analytics domain Trends in Analytics domain Trends in Analytics domain Trends in Analytics domain