Lablab AI Agents Hackathon, 13-15th Sept, 2024
Aim: Build an agentic workflow for the software development life cycle (SDLC)
Approach
- Build a PoC in a sandbox (dev) environment
- Break the workflow into four separate tasks, and use one agent per task
- Use crew.ai for the agents and llama 3.1 or Gemma-2-7b for the llms
- Use colab notebooks for the coding
- Use other tools as necessary, e.g., agentops, autogen, mindsdb, upstage, langgraph, composio, etc.
Agents
- Requirements Agent: Understand requirements from a given requirements doc
- Design Agent: Create a high level design doc
- Software Development Agent: Generate codebase to build a PoC (small project)
- Code Test Agent: Generate code tests
Other ideas: Agents for the MLOps life cycle
Workflow Steps
a. Requirements Gathering
- Task: Extract key requirements from a document
- Goal: Create a concise summary of the system's required features
- Outcome: Define the project scope (e.g., authentication, operations, task management, reporting)
b. High-Level System Design
- Task: Design the architecture of the system
- Diagrams Generated:
- Use Case Diagram
- Class Diagram
- Entity-Relationship Diagram (ERD)
- UI/UX Design for Dashboard
- Outcome: A document detailing the description, architecture and components of the system. It will include text and diagrams.
c. Code Generation
- Task: Develop the Code for the system
- Goal: Create functional code that implements core features
- Outcome: Working code implementing the basic system including authentication, database operations, and reporting
d. Code Testing
- Task: Run test cases to verify code functionality
- Goal: Ensure the system meets the requirements and works as expected
- Outcome: A detailed test report highlighting results and potential issues
Project Page
Colab Notebooks
- https://colab.research.google.com/drive/1vsYzaJSXERfU-uEXuK2aD4HA9QbEIVUl?usp=sharing
- https://colab.research.google.com/drive/1j9OTh4ridFnq6XQReiYcItzG9yi3Ydwh?usp=sharing
Future Work
- Improvements in Design Diagrams: Explore more AI-driven tools for automated generation of detailed design diagrams
- Customization: Enable more advanced configurations for tasks such as adding new agents or expanding the functionality
- Agent Testing: Add test, debug and monitoring features to our platform, such as AgentOps.ai
- Deployment: Plan for deployment of the final PoC in a production environment
References
- Developing a Multi-Agent System with CrewAI, https://lablab.ai/t/crewai-multi-agent-system,
- Mastering AI Agent Management with AgentOps: An In-Depth Guide, https://lablab.ai/t/agentops-tutorial