I’ll tell the story of how I solved the chaos in my Jupyter reports and built juport (Jupyter Report System) GitHub link. I’ll also share a few thoughts on how development is changing in the era of AI assistants.
The Problem: 100,500 Reports and No Order
I had accumulated a huge number of reports made in Jupyter Lab. Each one was a separate file with code, passwords, and all the behind-the-scenes “kitchen.”
Main pain points:
- Security. You can’t just share a report with management or accounting because it contains database access and code.
- Routine. There’s no centralized place to run reports, set up schedules, and have a unified viewing interface.
- Chaos. All results are scattered across folders, and finding the right Excel file took a lot of digging.
The Concept
I needed something that would let me develop reports in the familiar Jupyter Lab, then automatically run them, generate clean HTML versions without code, and gather all artifacts in one place.
The Solution: juport — a report management system
I created a system (well, more like vibe-coded it) consisting of two components:
- Jupyter Lab sidecar. A standard Jupyter Lab in a Docker container. This is where developers write and test reports as usual.
- juport — the management system. A Python web app that scans the notebook folder. It lets you run reports manually or on a schedule, executes them in an isolated environment, generates HTML versions stripped of extra info, collects all artifacts (Excel, images) into a single table, and provides a convenient web interface. Authentication is via LDAP.
How It Works
Report development:
- You create a notebook in Jupyter Lab.
- You write code, test it, save it.
- You use environment variables for configuration so passwords aren’t stored in code.
Report execution:
- You open the juport web interface.
- You see a list of all notebooks.
- You click “Run” or set up a schedule.
- The system executes the notebook and collects results.
Outcome:
- A clean HTML report without code or passwords, available for viewing.
- All Excel files, images, and PDFs collected in one place.
- A convenient download interface.
- Execution history and logs.
How It Was Built
I didn’t write a single line of code myself. I vibe-coded everything via Cursor using prompts.
Yes, exactly. Get used to it. That’s the reality.
Andrej Karpathy said that development will look very different soon. He’s right.
We, millennials, are the only generation that knew how to build a computer from scratch. Boomers were before the PC boom, and zoomers were born when everything was already set. The same is happening with code. In N years, experienced developers will get excellent results through prompts because they have 20 years of experience. That experience isn’t syntax knowledge, but understanding of:
- Architectural patterns
- Design principles
- Trade-offs
- Potential pitfalls
That’s why those who “get it” will get great results, and those who don’t will end up with a “brown substance.”
AI assistants aren’t a replacement for developers, but a tool that multiplies our speed. Experience and architectural understanding become even more important. And it will be harder for newcomers, because they’ll have to study technologies deliberately to get high-quality output from neural networks.
Takeaways
- AI assistants are not a replacement, but an instrument.
- Experience and understanding of architecture are critically important.
- Development speed for experienced specialists will increase many times over.
- Beginners will have to put in more effort to master the profession.