Hi Devs! 👋 I recently launched DocBlitz, a SaaS tool that auto-generates cloud documentation for your codebase and keeps it in sync — with AI-powered search and chat built-in. This post is a quick peek at why I built it, what I learned, and how I hope it helps others.
The Problem That Sparked It 💡
Like many developers, I’ve worked on projects with outdated, missing, or scattered documentation. It’s frustrating — for the dev team, for product managers, for support staff. And I’m not alone.
In fact, a 2021 GitHub Open Source Survey found that 93% of developers say incomplete or outdated documentation is a pervasive problem.
For early-stage startups or indie projects, it’s even worse:
- Docs often fall behind or never get written.
- Different developers write in different styles with different levels of quality.
- Teams forget how things work.
- New hires ask the same questions.
- Non-technical teammates feel lost.
I kept thinking: why doesn’t documentation just generate itself… and stay up to date?
The Solution I Built 🛠️
I built DocBlitz to solve that. It:
- Scans your connected repo(s) and generates a full-featured documentation site — in minutes.
- Keeps your docs in sync as your code changes.
- Lets anyone “chat with your docs”, using an AI-powered interface that understands your code.
It’s designed for small dev teams, startups, development agencies and solo devs who don’t have time to write (or maintain) docs but want something that looks and works great.
Here's a sneak peek at the output (a pretty docsite for a NetFlix clone I built while learning Next.js) 👇
🧠 What I Learned Along the Way
Building DocBlitz has taught me a lot.
I began the project with only a minimal understanding of AI technologies and almost no practical experience in building AI applications, so this was a significant personal challenge.
A few key takeaways:
LLMs don’t “understand” language the way humans do — they predict it
Large Language Models are trained to guess the next word in a sentence based on patterns in massive amounts of text data. It’s like a supercharged version of autocomplete. What’s wild is that by just learning to predict the next word really well, LLMs end up being able to write essays, debug code, translate languages, answer questions, and more—even though they don’t truly “know” anything. But if they don't “know” anything, how can they document an entire codebase, where every minor detail matters and there's no room for mistakes or guesses? Well, it turns out that LLMs are actually really good at this!
Through a combination of pattern recognition, token prediction, context (more on that below) and self-attention, LLMs can identify structure, infer intent, and generate documentation that feels surprisingly accurate—even for unfamiliar code. They’ve effectively “read” millions of codebases and documentation styles, so when they see a new function or class, they can often recognise patterns and produce descriptions that mirror what a human would write. They pick up on naming conventions, logic flow, and relationships between files and components, and use that to generate contextually relevant explanations.
This blew my mind the first time I tried it. A few lines of code in, and the LLM was already writing documentation that I would’ve otherwise spent hours drafting myself. Of course, it’s not perfect—LLMs still need guardrails, and human review is essential, especially for edge cases. But used thoughtfully, they unlock a new kind of development workflow: one where you can go from raw code to a polished, browsable documentation site in minutes, not days.
That’s the magic DocBlitz is built on.
Perseverance pays off
When it comes to understanding an entire codebase, across multiple repositories, context is a real problem. I noticed that even Cursor and VS Code struggled with this. They only ever 'see' a narrow area of context, which changes frequently as it doesn't remain in memory, and neither editor was able to connect the dots seamlessly between my separate front-end and back-end repos.
I was stuck for weeks on this, spending a small fortune on LLM tokens. I honestly came close to giving up. Thankfully though, I did have a breakthrough. It required designing a whole new system to parse and ultimately 'understand' the relationships and functionality between and within the project's directories and files.
You can imagine my delight when I ran a test on my new system and it worked! #staystrong
The power of queues
In designing the system behind DocBlitz, I realised there were many potential bottlenecks.
For example, an API rate limit on an LLM call. I'd frequently hit the limit as the app tries to process an absolutely massive amount of data in the shortest possible time. So what happens when I get an error code back rather than the response I need to move on. How could I handle this gracefully and ideally, retry this job later?
This is where queues come in. Queue based architecture is perfect here because whenever my app hits the rate limit on an API call now, I can simply add the job to a queue. My queue manager uses a cron job to pull tasks down from the queue at any interval I choose, and attempt to process it again. If it succeeds this time, it comes off the queue. If it fails again, it simply goes back into the queue. Nice!
What Makes DocBlitz Different?
There are some great tools out there — Docusaurus, GitBook, ReadMe — but I wanted documentation that:
✅ Writes itself
✅ Stays updated automatically
✅ Feels modern (think AI chat + full-text search)
✅ Works for both devs and non-dev teammates
✅ Deploys instantly to a hosted cloud docsite
Launching Today 🚀 (Free for Early Users!)
I’m launching the public beta now, and I’d love feedback from real devs using it on real projects.
👉 Try it today at DocBlitz.io
Works with GitHub. I'm giving away a free month (and maybe longer) for early adopters who are happy to provide feedback. Drop me a line at [email protected] for your free access code.
If you’ve ever struggled with documentation, I’d love to hear from you — comment below or DM me.
Thanks for reading! ✨
P.S. If you're a Dev.to writer or content creator curious to try it out or review it, let’s chat! I’d love to offer you early access.
