Orchestrating LLMs: Building a Python Bookkeeping System with AI Collaboration
In my recent project, I embarked on an exciting journey to develop a bookkeeping system using Python, heavily leveraging the capabilities of a large language model (LLM). The goal was to build an application and explore how LLMs can assume various roles in software development—acting as business analysts, system designers, and code developers.
The process began by engaging the LLM to create user stories and tasks and identify their dependencies. The LLM provided detailed tables outlining these dependencies, which I manually inputted into GitHub’s project board. Although the automation wasn’t complete, the collaboration was seamless. I structured the project board with a simple workflow:
- User Story
- Backlog
- To-Do
- In Progress
- Testing
- Done
This was supplemented with labels and milestones—all guided by the LLM’s output.
For each user story, I asked the LLM to generate tasks, and I requested developer instructions for each task. Then, it produced code snippets by prompting the LLM to act as a developer. I was an AI Orchestrator, coordinating this process—copying code into VSCode, testing it using scripts generated by the LLM, and troubleshooting errors with its assistance.
The system features APIs and dummy machine-learning plugins, hinting at future integrations. The LLM also helped generate comprehensive API and plugin user guides, covering the application’s design, implementation, and testing phases.
Throughout this experiment, I tested several LLM models from different providers. While some fell short, one stood out in delivering exceptional results. Although I prefer not to advertise it openly, I’m open to sharing details collaboratively. This project underscores the potential of LLMs in software development and the evolving role of developers in orchestrating AI capabilities.
I’m excited about where this synergy between human coordination and AI can lead, especially with plans to enhance the ML plugins. This approach doesn’t replace developers but augments our ability to build complex systems efficiently.
You can explore the project’s code on my GitHub repository, linked to my LinkedIn profile. I’m sharing this experience through my startup, RobotFace AI, as a testament to the innovative possibilities when humans and AI collaborate.