aiaicaramba / press-to-talk

Conversational AI for research journalists

Press to Talk: Voice-Driven Insights into Civic Data

Introduction

"Press to Talk" revolutionizes the way journalists interact with extensive archives of municipal council meetings. By leveraging advanced Conversational AI, this project allows journalists to use voice commands to query, navigate, and extract insights from vast datasets of council transcripts. Designed for simplicity and efficiency, "Press to Talk" bridges the gap between traditional research methods and the intuitive, conversational access to information.

Features

• Voice-Activated Searches: Perform searches using natural language through your microphone.
• Intelligent Query Processing: Our AI understands complex queries and delivers precise results.
• Interactive Feedback Loop: Refine your search based on the AI's questions to narrow down the exact information you need.

Contributing

We warmly welcome contributions from the community. If you're interested in enhancing "Press to Talk" with new features, improving the codebase, or fixing bugs, please check out our CONTRIBUTING.md for guidelines on how to get involved.

Contact Information

For any inquiries or further information, feel free to contact Leon van Bokhorst and Coen Crombach at Fontys University of Applied Sciences - FICT, Eindhoven. Your feedback is invaluable in making "Press to Talk" a better tool for journalists everywhere.

press-to-talk's Issues

Build scenario how to talk to council transcripts

Simple "Hero's Journey" on how a research journalist experiences the process of conducting research by talking to the crew of agents

• describe basic scenario (e.g., journalists looking for answers based on current events)
• identify agents (e.g., researcher, translator, name extractor, expander, communicator)
• identify tasks (e.g., selecting transcripts, formulate follow-up questions, answer, call, response)
• identify tools (e.g., context expander, search, summarizer, translator, name extraction)
• identify processes (e.g., serial, parallel, clamping, freedom, resulting format)
• visualize crew and process (easier to get feedback)

CrewAI PoC

Understanding the fundamental concepts of the CrewAI platform. Create a Proof of Concept to determine whether it serves as a foundation for the project to expand upon.

starting with crewapi

Starting with the 'Leon'-example, build from there.

Get in touch with project / domain owner

[PoC] How to use active inference to evolve into a reflection guide

PoC of a Multi-Agent System (MAS) with agents using Bayesian methods and Markov decision processes to facilitate reflective conversations with CrewAI as MAS platform.

• what's the goal?
• design agent structure
• design separate agents
• implement tasks

Feedback Ownership MAS

A feedback session between student and coach is recorded, transcribed and fed to the MAS (crewai based). Thereafter, while using FeedPulse, a conversational AI talks to the student about the session with the sole goal of enhancing their ownership of the feedback by asking questions and nudging to take responsibility.

• check if comparative feedback is viable as an option too?
• https://arxiv.org/abs/2302.03805

Documentation mkdocs and Obsidian integration

Create compelling contribution markdown file

Explain how to engage with the project.

Declaratie Coen en Leon