Twill Brain

Data Insights

3 min read

Data Insights

Date: 2026-01-21 · Duration: 84.26000213623047 min Organizer: Nader Attendees: Nader, Ahmad, Mike, Martin

Summary

  • AI Agent Implementation: Chat AI will assist sales reps live, using WebSocket for real-time updates and simplified responses.
  • Voice AI Functionality: Voice AI to transcribe calls and provide insights, improving sales efficiency with role-based access for security.
  • Commission Estimation Phases: Project will analyze data from 43 merchants, using features like transaction values and refund rates for modeling.
  • UI Commission Display: UI will show commission estimates as KPI tiles, with options for last month and month-to-date views.
  • Data Strategy for AI: Rolled-out data access will enhance AI performance, focusing on transaction details and role-specific queries for accuracy.
  • Modular ML Pipeline: Commission model will use various regression techniques and focus on robust evaluation methods for ongoing refinement.

Action Items

  • Ahmadreza Abdoli Filter loaded merchant data to include only those with commission data (49:30) Implement data quality handling, setting null values to zero in commission estimates (50:00) Perform exploratory data analysis including distribution and bivariate analysis of commissions (50:30) Proceed with feature engineering incorporating new features such as average transaction value, refund rates, fee rates (51:45) Develop linear regression model with train-test split (80/20) for initial commission prediction and evaluate per merchant and aggregated (01:02:00) Create a markdown documentation with clear tables and pipelines for model training and evaluation to allow cursor-based batch processing (01:16:30) Explore and implement advanced regression models including tree-based models, hyperparameter tuning, and model comparison after initial implementation (01:19:30) Share the markdown document and updates with team for review (01:23:40)

Martin Elias Costa Provide detailed ticket for websocket integration and assign to responsible developer (22:55) Refine chat UI design to reflect minimalistic AI bot intervention with streamlined, concise responses (16:20) Share existing residual report data and monthly aggregated features to assist commission model development (52:40) Review and provide feedback on commission estimation phases and assist in feature engineering (53:30)

Nader Atrchin Oversee AI agent architecture design including authentication approach for chat and voice AI agents, ensuring role-based access (01:20) Coordinate between teams on voice and chat AI toolsets and pipeline integration (37:00) Drive discussion on live websocket architecture to support real-time UI updates from AI and human interactions (22:15) Confirm AI’s listening integration approach on phone calls and interface for sales reps including three-way calls and merchant identification (43:00) Define UX expectations for commission estimates and coordinate integration with Ahmed’s work (01:09:10) Review markdown and ML pipeline documentation for commission modeling, provide detailed feedback and guide model evaluation strategy (01:22:30)

Michael Wright Provide research data on POS terminals, integration interoperability, and merchant questions frequently asked to inform AI knowledge base (27:15) Share pricing and fee schedule data for installed processors (EMS, Lucron) relevant to commissions modeling (01:06:20) Advise on practical user workflow scenarios for AI agents in call and chat environments (43:00)

Graph View

Backlinks