<aside> 🏎️

Go to the Quick Start Guide!

</aside>

Dev Lead: @Jakub Gil

Wiki: 42 Hackathon - Conversation Analytics Wiki

Front end application: https://conversation-detection.stage.cere.io/

Git repository: https://github.com/cere-io/42-hackathon-nlp-llm-conversation-analytics

Tech Summary:

Objective

This CEF.ai Prototype is an AI-powered conversation analysis system designed to process, structure, and analyze large-scale messaging data from communication platforms like Whatsapp groups, and Telegram groups.

The goal is to transform community discussions from a sea of messages into clear, actionable insights, e.g. what a community is talking about, who are key contributors, and what topics drive the most engagement - all in (near) real-time.

Goal/Values For Hackathon:

• Extract insights from collective conversations by combining intelligent real-time data segmentation and NLP models.
• Key actions/learnings for each participant/student
  • Quickly build/customize/launch open-source models to work with dynamic datasets while conversations are happening
  • Simulate how our prototypical real-time data engineering platform can help you to compare/evaluate LLM results dynamically to understand relevance and content within a temporary and extended time context window
  • Continuously build up and archive the conversational insights in the format of vectorized conversation/thread mapping
• Your objective is to understand how to use different model and tuning techniques, and perhaps even more advanced data segmentation techniques to achieve better score/outcome than the provided sample

🕵️ Agent Deploy/Test/Experimentation Execution Logs

• 🎯 Deliverables 2024-02-26 (Hackathon Day 1)
• 🎯 Deliverables 2024-02-27 (Hackathon Day 2)
• 🎯 Deliverables 2024-02-28 (Hackathon Day 3)
• Some Additional Tips and Guidance Experiment & commit to existing modules (namely: Pre-grouping, LLM inference) for best performance on evaluation data (across Spam Classification, Conversation Clustering, Topic Labeling)