ChatRnD

Current Version: 0.0.1

A Multi-Agent Framework for Scientific Research and Development (RnD), inspired by the ChatDev and CAMEL frameworks, an extension of the PhDBot.

Project Timeline

Version 1: Q3 2024 - Q4 2024

This project aims to develop a proof-of-concept for an upcoming grant proposal to-be-submitted to ResearchHub for Communicative Agents in the University Research Use Case. This project is inspired by the call-for-action described Here: ResearchHub Bounties.

ChatRnD will specialize in creating a multi-agent or team version of PhDBot.

A. Top 5 Priorities for Development

1. Define role structures for ChatRnD and create a new architectural blueprint.
2. Design the ChatEnv configuration and base/background system prompts.
3. Redefine ChatConfig structure and content based on roles in ChatRnD.
4. Develop basic task parsing and assignment upon receiving a user prompt for a "Research Topic"; define task termination criteria.
5. Create Wireframe design for all reports type. Test the generation (exploratory) reports based on literature findings and analyses within the research topic.

B. Development Guideline

We envision ChatRnD to be fully open-source, and thus would be constructed with complete support for modularity in choosing LLM-Engine and DataHub Framework. A Bring your Own (LLM) Model and Tools are also highly encouraged. Thus, we will strive for ease of import and compatibility with most commonly used framework/tools out there. This will be updated from time to time based on user-request and development priorities.

C. Goals

Tier I: Simple

• Objective: [Exploratory Research Report] Provide a 5-8 page weekly/monthly research report on a specific topic equivalent to first/second year Graduate Student.
• Scope: Perform literature reviews, ideate possible improvements, assess feasibility, etc.
• Duration: Complete the exploratory research report within 30 minutes of receiving the research topic prompt.

Tier II: Moderate

• Objective: [PhD Qualification Proposal] Deliver a 10-25 page detailed analysis, similar to what to be expected of in PhD Qualification Proposal.
• Scope:
  1. Detailed Background/Significance for the research
  2. ChatRnD should be capable of deriving/outlining the relevant Theory/Methods underlying the topic
  3. Produce a Specific and Feasible Research Experiment Plan as necessary, including basic settings and pro/cons when appropriate.
  4. ChatRnD must be able to provide a mock of expected research outcome, both in the context of positive or negative outcome.
  5. On top of 1-4, ChatRnD must be capable of enlist and describe the limitations/alternative approaches to what laid out in the Report.
• Duration: This whole process from receiving Research Topic prompt (in which intermediate Exploratory Research Report will be produced) to finishing the PhD Qualification Proposal should be <3 hrs of continuous operation/revisions. Termination will be imposed either if "improvement" is deemed to be negligible or the termination considerations were crossed.

Tier III: Complex

• Objective: [Final RnD Report] Prepare a reinforcement analysis based on the proposed PhD qualification proposal, including metadata analysis from research papers and in-house implementation of suggested methods.
• Scope: The goal for Tier III is for ChatRnD to be able preparing a reinforcement analysis (e.g., doing calculation, using external tools) based on the proposed PhD Qualification Proposal. The Analysis will include a meta-data analysis from research paper and/or some in-house implementation of the method suggested. Moreover, after completion of the Final RnD Report, ChatRnD should be integrate it's collective memory/knowledge to a PhDBot representative that will eventually present the report to human user. The presentation will be mainly a text-to-speech interaction between user and the PhDBot being "Examined".

Ideally if the human user(s) satisfied with the explanation given, the PhDBot will then be granted full PhDBot Candidacy status.

• Duration: The whole process from receiving Research Topic prompt to Final RnD Report (including all the intermediate output in Tier I and Tier II) is expected to be done within <8 hrs

D. Deliverables

• Public code repositories on GitHub.
• A main info page at ChatRnD.
• An interactive app for proof-of-concept at ChatRnD App.
• A gallery showcasing case studies at ChatRnD Gallery.
• Options for self-hosted servers for user testing, development, and deployment.

E. Future Stretch Goals

Being able to work in a continuous way toward a full-fledge long-form PhD dissertation something about at least 50-100 pages long. The document will highlight a void/missing link in theory/method/concept within a particular research field, and work out possible novel approach or solution toward addressing it. A decent work that demonstrate how the proposed approach in action should be included. The expected timeline will be for the ChatRnD to run continuously for several days up to a week or so. To then come up with the first version of the dissertation.

Special Thanks:

• @Tyler Diorio (Chief of Staff at ResearchHub)
• @Jeffrey Koury (Director at ResearchHub Foundation)

These individuals have initiated discussions and brainstorming sessions that led to the start of this open & collaborative decentralized science (DeSci) project.

Call for Contributors: After the initial framework for this project is worked out we are inclined to open a call for contributors to develop scientific use-cases and training the Agents on different expertise as part of our ResearchPreneur Partnership Program (to be announced)

By DreamBrook Labs
June 2024