FDAi Autonomous Project Manager AI Agent for Repository Management

[mikepsinn]

We want an AI agent, named FDAi, that will eventually serve as the director of the FDAi project. This AI agent will be responsible for automating various repository management tasks, significantly improving efficiency and ensuring consistent project progression. The ultimate goal is to create a self-improving system that will autonomously manage the repository, contributing to the global mission of accelerating clinical research and maximizing health and happiness.

Objectives

1. Automate Issue Labeling: The FDAi AI agent must automatically label incoming issues based on their content, using a predefined set of labels for categorization.
2. Milestone Creation and Management: The agent should be capable of creating and assigning milestones to issues and pull requests, facilitating better tracking of project progress.
3. Issue and Pull Request Management: FDAi should autonomously manage issues and pull requests, including the ability to create, update, and close them based on project requirements and contributions.
4. Data Indexing and Analysis: It should index and analyze the contents of the FDAi GitHub repository, extracting valuable insights to guide project direction and priorities.
5. Community Engagement: The agent must foster community engagement by facilitating discussions, responding to queries, and encouraging contributions from developers and researchers.

Requirements

• Integration Capability: Ability to integrate with the GitHub API for performing various repository management tasks.
• Self-Improvement Mechanism: The agent should have the capability for self-assessment and self-improvement, learning from its actions to optimize its performance over time.
• Security and Privacy: Ensure strict adherence to security and privacy standards, safeguarding project data and contributor information.
• Open-Source Development: Development should be carried out in an open-source manner, encouraging collaboration and transparency throughout the project lifecycle.

Milestones

1. Research and Planning: Conduct thorough research to identify the best approaches and technologies for developing the FDAi AI agent. (Duration: 1 Month)
2. Prototype Development: Develop a prototype that demonstrates basic capabilities such as issue labeling and milestone management. (Duration: 3 Months)
3. Testing and Feedback: Test the prototype within the community, gathering feedback for improvements. (Duration: 2 Months)
4. Full-Scale Development: Implement feedback and develop the full-scale AI agent with complete functionalities. (Duration: 6 Months)
5. Deployment and Monitoring: Deploy the FDAi AI agent and monitor its performance, making adjustments as necessary. (Duration: Ongoing)

Contribution Guidelines

• Code Contributions: Please submit pull requests with detailed descriptions of your contributions and how they enhance the FDAi AI agent's functionality.
• Feedback and Suggestions: We welcome feedback and suggestions through issue submissions. Please label them appropriately as feedback or suggestion.
• Community Discussion: Engage with the community through discussions to share ideas and collaborate on enhancing the FDAi AI agent's development.

[supmo668]

The service will be hosted from the following repository:
https://github.com/supermomo668/FDAi-helper-agents
The API service will be hosted either on a VM in GCP or with Cloudrun. Will be posted here momentarily

[supmo668]

Service will be hosted statically here, but am debugging container deployment issues:
https://direct-lacewing-merry.ngrok-free.app/docs#/default/query_llama_index_query__get
Just send text to /query/ to ask about the FDAi repo.
[TODO]
More testing about the quality of the response. Takes around ~10 minutes for to index all the files.
If the pipeline do the rag decently ok, then will automate in Cloudrun and the service should be ready in production. Also exchanging correct API tokens

[mikepsinn]

1st Milestone: https://chat.fdai.earth page where you can talk to the markdown contents of the repo

Proposed Implementation Steps:

• Deploy https://khoj.dev/ somewhere
• Index https://github.com/FDA-AI/FDAi
• Add credits link to https://khoj.dev/ at the bottom of the pages
• Add a link to the Swagger API docs in this issue
• Make a JS script we can include on any website with an Intercom-like chat widget using an existing framework or ask GPT to make one. Connect it to the API. Example in the lower right corner at https://studies.fdai.earth/variables

Once that's done, we can think about the best way to create an agent to respond to GitHub issues and pull requests with useful information. Then, eventually, we can have the agent create issues and plan milestones. And then hopefully be able to self-improve soon after.

Also, try applying to
https://aistudio.google.com/app/waitlist/97445851?utm_source=agd&utm_medium=referral&utm_campaign=blog-feb&utm_content=
to get 1M token context.

[supmo668]

test Khoj server is hosted up here : https://certainly-inspired-duck.ngrok-free.app/, looking into programmatic approach for other actions.

[mikepsinn]

Thanks, @supermomo668

Do you know the URL for the Swagger API docs?

I guess maybe we don't need to modify that project, and we could just use the hosted version, get an API key, and create a script to add the chat bubble.

[supmo668]

Have some issue connecting the repo on the self-hosted app. Anyone had the same issue?
https://certainly-inspired-duck.ngrok-free.app/config
The API is definitely somewhere on the server. The app is built in Django, so I can dig into it but there're no terms found searching the codebase for Swagger or OpenAPI

[mikepsinn]

https://github.com/search?q=repo%3Akhoj-ai%2Fkhoj%20swagger&type=code

[supmo668]

Swagger Docs: https://certainly-inspired-duck.ngrok-free.app/docs