This is a submission for the Built with Google Gemini: Writing Challenge
What I Built with Google Gemini
I built Archlyze, a browser-based AI code analyzer that lets you paste in source code and get an instant, structured analysis: what the code does, potential issues, security concerns, and architectural patterns.
The entire thing runs without a traditional backend. Gemini is the backend. The API is called directly from the browser, which surprised me with how capable and practical that turned out to be for a tool like this. It felt like a different mental model, instead of building a server to orchestrate AI calls, I just... talked directly to the model from the client.
Archlyze is also getting a significant update today! with an Executive Briefing feature, a full on-demand presentation layer powered by Gemini that transforms raw code analysis into a management-ready deck. It includes:
- Executive Translation: Gemini rewrites the analysis in plain, jargon-free language for non-technical stakeholders
- Code Walkthrough: Side-by-side view with full source and plain English explanations, section by section
- Architecture Diagram: Auto-generated Mermaid.js flowchart rendered live in the presentation
- Security Scorecard: Visual risk heatmap with a 0–100 score and per-component health table
- Multilingual Briefings: Configurable output language: English, Norwegian, Japanese, or others
- Export: Download as Markdown, PowerPoint (.pptx), or PDF
- Editable: Toggle edit mode to tweak the Markdown before exporting
It runs entirely on-demand and doesn't touch the standard analysis pipeline.
Demo
👉 Read my original build post here
What I Learned
The biggest lesson was about workflow switching. I started the project entirely in Google AI Studio, iterating on the React setup through a lot of trial and error, probably 10+ attempts before things clicked. That repetition wasn't wasted though; years of working with Cursor taught me that reiteration is just part of the process, not a sign something is wrong.
At a certain point, when the project had good momentum but was growing in complexity, I switched to Cursor. Not because AI Studio failed me, but because I needed more control over session state and context. Cursor gives me tighter, more predictable context windows and better tooling for managing a growing codebase. The two tools ended up being complementary rather than competing: AI Studio for exploration and prototyping, Cursor for refinement and complexity management.
The other surprise was how naturally Gemini fits as a "zero-backend" architecture. For the right kind of project, especially developer tools and internal utilities, calling the model directly from the client is a completely legitimate approach, not a hack.
Google Gemini Feedback
Honestly impressed, especially early on. Gemini was fast, coherent, and produced genuinely useful output right out of the gate for a project like this.
Where I ran into friction: context management over long sessions. AI Studio feels designed around a long, continuous conversation, and that works well up to a point. But as the project grew and my prompting got more layered and less structured, I started losing confidence in whether the model was still tracking the full picture. I'm not sure if it was the model, my prompting style, or the lack of a rules/system-prompt setup, but it felt like complexity was the ceiling.
For focused, well-scoped tasks? Excellent. For managing a multi-session project with evolving requirements? I'd want more control, custom instructions, explicit context resets, and better visibility into what the model is "holding." That's ultimately why I moved to Cursor for the harder parts.
But the core capability is strong. The speed and quality on the initial build genuinely impressed me, and using it as a live backend for a browser app is an approach I'll absolutely use again when it fits.
Top comments (0)