Modular code generation

[4Rom1]

ModularCodeGen Crew

Welcome to the ModularCodeGen Crew project, powered by crewAI.

🧠 CrewAI – Modular Engineering Team (Agentic Version)

This project is an agentic and modular transformation of the original engineering_team project.
Unlike the original monolithic implementation, this version introduces fully separated backend modules, each with its own unit tests and automated frontend integration, built using an orchestrated multi-agent workflow defined in tasks.yaml.

---

🚀 Overview

The CrewAI system automates the design, coding, testing, and integration of a Python backend and frontend architecture using specialized AI agents.
All modules and their test files are stored under account_directory/.
After running all tasks, the generated structure looks like:

account_directory/ ├── module_a.py ├── test_module_a.py ├── module_b.py ├── test_module_b.py ├── ... ├── accounts.py # main frontend API ├── test_accounts.py # frontend headless test └── README.md

The main API can be launched with:

$ uv run python accounts.py

⚙️ Architecture Summary

This system decomposes the engineering workflow into five coordinated tasks, each handled by a specific agent role.
All task definitions are provided in tasks.yaml.

---

🧩 1. design_task

• Agent: engineering_lead
• Goal: Analyze given requirements and design a modular backend.
• Output:
  • List of module names (module_names)
  • Specification of classes, methods, and interactions (module_specs)
  • JSON schema describing the architecture
• Key Point: Each module must have a unique name.

---

💻 2. code_task

• Agent: backend_engineer
• Goal: Implement Python code for each designed module.
• Output:
  • Source files (module_sources)
  • Corresponding test files (module_tests)
• Test Framework: unittest
• Constraint:
  • Outputs must be pure Python code (no markdown formatting).
  • The number and names of modules must match the design_task exactly.
  • Tests run in a dedicated environment.

---

🗂️ 3. writer_task_back

• Agent: file_writer
• Goal: Persist all generated backend modules and tests to disk.
• Output Directory: account_directory/
• Output Files:
  • {module_name}.py
  • test_{module_name}.py
• Also Generates:
  • A README.md describing each module and its functionality.
• Result: All backend code is saved under account_directory/ and ready for import.

---

🌐 4. frontend_task

• Agent: frontend_engineer
• Goal: Build a lightweight Python API using gradio to expose endpoints for all backend modules.
• Outputs:
  • fronted_source – the frontend API source code.
  • fronted_test – test verifying headless initialization of the API tool.
• Behavior:
  • The frontend automatically imports all backend modules.
  • It runs in headless mode to confirm proper initialization.

---

💾 5. writer_task_front

• Agent: file_writer
• Goal: Persist frontend code and test files.
• Output Directory: account_directory/
• Files Written:
  • accounts.py
  • test_accounts.py
• Note:
  Before generating the frontend, the backend modules are copied into the same directory so the frontend can import and test them in headless mode.

[ed-donner]

Hey - I really appreciate this, but with 23 files and 4.4k lines of code, this is on the large side for the repo.. Please could you either:

• Cut back the number of README (+emojis!) and test files - focus on the example to be shared
• Alternatively, replace this with a single md that describes your project and links to the implementation in your repo I hope that makes sense.. please see Q3 here for more: https://edwarddonner.com/pr

[rgarnier-r7]

Hey - I really appreciate this, but with 23 files and 4.4k lines of code, this is on the large side for the repo.. Please could you either:

• Cut back the number of README (+emojis!) and test files - focus on the example to be shared
• Alternatively, replace this with a single md that describes your project and links to the implementation in your repo I hope that makes sense.. please see Q3 here for more: https://edwarddonner.com/pr

Thanks for your feedbacks! I have removed the test files, one of the README, the emojis on the remaining README. Let me know if I missed something.