Originally, the plan was to use Astra 6 and have qwen do the work, which is the standard approach. However, qwen wasn't able to follow Astra 6's instructions well, and it was too wasteful to see Astra tokens melting away just for review. So I changed the working method.
In other words, qwen outsources work to Astra
Codex monitors qwen's results through an API ㅋㅋ It took 26 hours (waiting time for other resources was about 16 hours)
Qwen27B+Astra(Web) App Factory to Make a Black Hole Simulator
What was the process?
If you explain the app you want to make to Qwen-Astra, Qwen will organize the requirements and request coding outsourcing to ChatGPT web.
ChatGPT 6 Pro used Three.js, WebGL2, and GLSL to write the core code and modification patches, and qwen collected and integrated the received code into the existing project.
The integrated result was then passed on to a separate VectorBuild task. VectorBuild handled actual file generation, type checking, testing, building, and code hash verification.
Finally, Codex and Vectorman monitored the entire progress, resuming failed tasks, restoring services, and verifying WebGL2 screen displays for desktops and mobiles.
Final result
Implementation of Kerr black hole and accretion disk, gravitational lensing, stellar collapse and supernova, event horizon formation, and spaceship black hole flyby
Estimated Token Usage for This First Development
Qwen 27B: Approximately 290,000~750,000 tokens
Used for requirement analysis, outsourcing request writing, code collection, conflict resolution, test result analysis, and recovery instructions.
ChatGPT 6 Pro: Approximately 70,000~130,000 tokens
Used for writing Three.js·WebGL2·GLSL code, implementing physics and screen logic, and writing error fix patches.
This time, we developed not only the black hole app but also the ChatGPT web connection, Qwen integrated structure, VectorBuild task generation, automatic/manual recovery, status observation, and mobile compatibility. Therefore, the token usage was large.
Estimated Token Usage After App Factory Stabilization
Qwen 27B: Approximately 100,000~250,000 tokens
Since the outsourcing·integration·verification structure is already complete, we can focus on requirement organization and code collection.
ChatGPT 6 Pro: Approximately 40,000~80,000 tokens
We will focus on writing the actual app code and necessary modifications without connection and recovery processes.
※ The token figures are estimated values calculated based on stored code volume and work records. They are not the exact usage amounts reflected in official billing receipts.
