Over the last weeks I decided to experiment with AI-assisted development workflows to see how much faster I could go from idea to working MVP.

Instead of treating AI as a “code generator”, I approached it more like an engineering accelerator:

• brainstorming partner


• rapid prototyping assistant


• UI iteration helper


• debugging companion


• architecture sounding board

The result was two small projects:

• Altivo — a lightweight portfolio tracker


• Hopway — a visual route planning app

Both projects are intentionally lightweight, but they helped me explore a modern development workflow using:

• Claude AI


• OpenCode AI


• DeepSeek V4 Flash


• Qwen 3.6 Plus

The Projects

Altivo

Altivo is a lightweight and privacy-friendly portfolio tracker focused on simplicity and static deployment.

Live Demo:

https://danieljvsa.github.io/altivo/

GitHub:

https://github.com/danieljvsa/altivo

The idea was to build something lightweight that could:

• track ETFs


• support crypto


• run entirely as a static frontend


• avoid unnecessary complexity

I wanted a cleaner and simpler experience than many traditional portfolio platforms.

The project also became an interesting exercise in:

• frontend architecture


• UI simplification


• API integration


• local-first persistence

Tech Stack

• HTML


• CSS


• JavaScript


• GitHub Pages


• Financial Market APIs

Hopway

Hopway is a lightweight route planning and travel exploration tool.

Live Demo:

https://danieljvsa.github.io/hopway/

GitHub:

https://github.com/danieljvsa/hopway

The concept behind Hopway was creating a map-first experience where users can:

• create routes


• visually explore destinations


• share travel plans


• experiment with AI-generated route ideas

The app uses CARTO maps and focuses on keeping the UX lightweight and visual.

I wanted it to feel more like an exploration tool than a traditional route planner.

Tech Stack

• HTML


• CSS


• JavaScript


• CARTO Maps


• GitHub Pages

The AI Workflow

One of the most interesting parts of these projects was experimenting with different AI tools together.

Each model had different strengths.

Claude AI

Claude was the strongest tool for:

• architecture discussions


• reasoning


• debugging


• planning features


• refining ideas


• writing structured prompts

It was especially useful when thinking through:

• application structure


• user experience


• roadmap decisions


• feature tradeoffs

OpenCode AI

OpenCode became the main execution layer for applying changes directly inside the repositories.

This made iteration significantly faster because I could:

• define tasks


• refine prompts


• generate implementations


• review results quickly

The workflow started feeling much closer to collaborating with a junior developer that could iterate extremely fast.

DeepSeek V4 Flash

DeepSeek V4 Flash was surprisingly useful for:

• quick fixes


• smaller implementations


• fast iterations


• UI experimentation


• alternative code suggestions

Because it was fast and free, it became useful for high-frequency iteration loops.

Qwen 3.6 Plus

Qwen 3.6 Plus was particularly good for:

• generating alternative approaches


• refining frontend ideas


• brainstorming UX improvements


• comparing implementation styles

Sometimes asking multiple models the same question produced surprisingly useful contrasts.

What AI Was Actually Good At

The biggest productivity gain was not “AI writing code”.

It was reducing friction.

AI was extremely useful for:

• generating boilerplate


• accelerating iteration


• refining UI ideas


• debugging faster


• exploring alternatives


• reducing context-switching

For small MVPs, that becomes incredibly powerful.

What Still Required Human Control

The important engineering decisions still required human oversight.

Things that still needed strong manual control:

• architecture decisions


• consistency


• product direction


• deployment


• performance considerations


• long-term maintainability

AI accelerated execution, but it did not replace engineering judgment.

Lessons Learned

A few things became very clear during these projects:

• Smaller focused prompts worked better than giant prompts


• AI performs best when architecture is already defined


• Iteration speed becomes one of the biggest advantages


• Reviewing generated code remains critical


• Combining multiple models produced better results than relying on a single one


• AI is especially powerful for reducing setup and boilerplate friction

One of the most interesting outcomes was how quickly experimentation became possible.

Normally many side projects die during the setup phase:

• too much boilerplate


• too much context switching


• too much repetitive work

AI dramatically reduced that friction.

That allowed me to spend more time on:

• product thinking


• UX refinement


• architecture


• experimentation

Instead of repetitive setup work.

Final Thoughts

These projects are small, but they represent something bigger:

a shift in how rapid MVP development is evolving.

AI-assisted workflows are becoming extremely valuable for:

• side projects


• prototypes


• internal tools


• experimentation


• product validation

AI did not replace development.

But it significantly reduced the time between:

idea → prototype → working application.

As a backend developer, these projects also helped me improve:

• frontend awareness


• UX thinking


• rapid delivery skills


• product-focused development

And honestly, the workflow was fun.

If you're experimenting with AI-assisted development workflows too, I'd love to hear how you're using them.

Links

Altivo

Live Demo:

https://danieljvsa.github.io/altivo/

GitHub:

https://github.com/danieljvsa/altivo

Hopway

Live Demo:

https://danieljvsa.github.io/hopway/

GitHub:

https://github.com/danieljvsa/hopway

About Me

Portfolio:

https://danieljvsa.vercel.app/

GitHub:

https://github.com/danieljvsa