Holy Typos, Batman! How I Built 'SpellJump'

Prakhar54-byte — Thu, 21 May 2026 16:32:37 +0000

Hey everyone! 👋

“Kabhi kabhi lagta hai apun hi bhagwan hai,” especially when your code compiles on the very first try. But then, you look at your docstrings, variable names, and READMEs, and realize they are absolutely full of spelling mistakes. Yep, that was me.

I was spending too much time manually hovering and clicking on those annoying red lines. I wanted a tool that didn't just passively highlight errors, but actively jumped to them like the Batmobile responding to a signal. So, I built SpellJump — an AI-powered typo jumper for VS Code and Cursor.

💡 How I Got the Idea Standard spell checkers feel slow and boring. I wanted a blazing-fast, offline-first tool that let me jump from typo to typo using a simple Ctrl+Shift+J shortcut. And because "Why so serious?", I decided to bake in some hidden DC Batman Easter Eggs. (Seriously, type "batman" or "gotham" in your editor with the extension on, and see what happens! 🤫)

I built the core extension using TypeScript and the VS Code Extension API. The biggest "Moye Moye" moment during development was handling real-time text document synchronization. You type fast, and the extension needs to parse the text, find typos, and update the Problems panel and wavy underlines instantly—all without lagging the editor. Figuring out the exact text-range mappings while maintaining peak performance was a solid headache.

🧠 Putting my Linux & AI/ML to Work For the core logic, I started with nspell for a fast, offline baseline. But I wanted to push the boundaries with AI. Here is how I utilized my ML and Linux environment for Phase 2:

I fired up my Linux terminal and wrote Python scripts to generate a synthetic dataset of 2,000+ common developer typos.
I trained a baseline model and then brought out the big guns: fine-tuning a DistilBERT model using PyTorch and Hugging Face.
I ran the training workloads in the background using good ol' nohup

(my train.log file was my best friend for days).
The real trick: You can't just run heavy PyTorch models inside VS Code without eating up all the RAM. So, I exported the trained model to an ONNX format (spelljump.onnx). Now, it runs locally, completely offline, and lightning-fast!
🚀 Deployment & The 10-Day Milestone When it came to publishing, I chose Open VSX—the open-source alternative to the Microsoft Marketplace that powers editors like Cursor and VSCodium.

Deploying was incredibly smooth. No massive corporate hoops to jump through. All I needed was a GitHub account, the ovsx CLI, and my .vsix package. One command (ovsx publish), and it was live.

And the response? In just 10 days of deployment, SpellJump crossed 185+ downloads! Seeing developers around the world actively use a tool I built from scratch is a feeling better than seeing the Bat-Signal light up the sky.

"It's not who I am underneath, but what I do that defines me." And right now, what I do is help you fix your typos faster. 😂

If you use VS Code or Antigravity , give it a spin, try to find the Easter eggs, and let me know your thoughts!

🔗 Check it out on Open VSX: https://open-vsx.org/extension/prakhar-iitj/spelljump#review-details 🔗 Source Code on GitHub: https://github.com/Prakhar54-byte/spelljump

VSCode #Cursor #TypeScript #MachineLearning #OpenSource #OpenVSX #DevOps #BuildInPublic #Batman #DeveloperTools

DEV Community: Prakhar54-byte

[Boost]