- AI-assisted coding made a complex image processing pipeline accessible to a developer with zero video production experience.
- The project shows AI-assisted coding can handle real stack decisions — the assistant consistently chose Python over JVM for good technical reasons.
- A five-stage pipeline handles everything from GPS filtering to neural image alignment, turning years of run photos into a single composite year.
- The alignment step — warping hundreds of photos to match a single reference frame — is where the real technical complexity lives.
- AI-assisted coding made a complex image processing pipeline accessible to a developer with zero video production experience.
- The project shows AI-assisted coding can handle real stack decisions — the assistant consistently chose Python over JVM for good technical reasons.
- A five-stage pipeline handles everything from GPS filtering to neural image alignment, turning years of run photos into a single composite year.
- The alignment step — warping hundreds of photos to match a single reference frame — is where the real technical complexity lives.
When AI-Assisted Coding Meets a Personal Creative Project
AI-assisted coding has been rewriting what’s possible for developers who sit somewhere between hobbyist and expert — and this story is a clean example of why. Nicolas Fränkel, a developer who publishes regularly at A Java Geek, spent years on his countryside runs photographing the same spot from the same rough angle. The quiet ambition behind those photos? Turn them into a time-lapse that captures one place shifting through every season. The problem was that Fränkel had no meaningful background in image manipulation, video encoding, or codec selection. He describes his practical knowledge of codecs as knowing when a movie file won’t play on his internet box. That gap — between a good idea and the technical execution — is exactly where modern LLM-based coding assistants have started earning serious credibility.
This isn’t a story about pushing the boundaries of machine learning research. It’s something more interesting: a real-world account of what happens when someone with domain knowledge but limited specialist skills hands a genuinely complex project to an AI assistant and tries to hold it accountable every step of the way.
The Stack Decision — and Why Python Won Every Time
One of the smartest things Fränkel did early on was treat the AI-assisted coding process like an actual engineering project. That meant taking the stack selection seriously rather than just accepting the first answer. He described the project concept to his assistant and asked it to choose between Python, the JVM, or something else entirely. The answer was Python. He pushed back multiple times across the project’s lifecycle, asking the assistant to reassess. It stuck with Python each time.
The reasoning is technically sound and worth understanding. Fränkel’s instinct was to argue for JVM performance — a reasonable assumption from someone who’s spent time in Java. The assistant’s rebuttal was direct:
Source: https://dev.to/nfrankel/seasons-time-lapse-the-foundations-4d4b
