Meet Vachan, a final-year engineering student from Davanagere, Karnataka, who is redefining how we interact with Artificial Intelligence. While the world is obsessed with generating flashy images from prompts, Vachan noticed a gap: current AI is a "black box" that starts from noise, making precise technical diagrams a nightmare to edit. Instead of settling for the status quo, he embarked on a self-driven experiment - "Draw AI".
This isn't just about making pictures; it’s about giving an LLM (Large Language Model) a virtual pencil and a blank canvas to draw exactly what you describe, line by line. From a Tier-2 city college to building complex Agentic AI workflows using LangChain, Vachan’s story is a testament to the power of self-learning and curiosity. He proves that you don't need a massive team or a Silicon Valley zip code to innovate, just a laptop, an internet connection, and the will to solve a problem.
Tell Us Something About Yourself ?
Hello! I am Vachan, based out of Bengaluru, Karnataka. My romance with computers started pretty early, probably when I was around 12 or 13 years old. By the time I was 15, I was deep into programming. Currently, I am in my final year at Bapuji Institute of Engineering and Technology in Davangere. What really fascinates me is control. I like to be in charge of what the computer is doing. I knew early on that my life would revolve around engineering, specifically computers or electronics.
What is the problem you are trying to solve with AI?
Here is the issue: current AI image generators (like Stable Diffusion) start with "noise" to create an image. It’s like a black box. If the AI draws a box in the wrong place, you cannot easily go in, click that box, and delete it without ruining the image. For engineers who need structured diagrams or flowcharts, this randomness is a headache. You get spelling mistakes or weird lines because the model doesn't truly "understand" the structure.
Tell Us the Story Behind Your Motivation with this AI Experiment?
I am a bit of a perfectionist. I wanted a way where, if the AI makes a mistake, I can manually fix it. I wanted the output to be an SVG (Scalable Vector Graphic) file, where every line and shape is editable. Since no existing LLM could generate these editable, precise technical drawings perfectly, I decided to build it myself. It was an experiment to see if I could make an AI that draws like a human starting with a blank sheet of paper.
Solution Overview - The AI Use Case
I call it "Draw AI". Imagine you are telling a friend how to make coffee. You would draw a box for "boil water," then an arrow to "add coffee," right?. My solution lets you talk to the AI, and the AI acts like an artist on an empty canvas. It draws lines, circles, and boxes in real-time based on your conversation. It’s perfect for flowcharts, system architecture diagrams, or even storyboards for films. It’s not just generating pixels; it’s generating logic.
Technical Details - How AI is Actually Utilised
This is where it gets geeky! I didn't train a new model; I used an Agentic Workflow using LangChain.
The Agent: I created an AI agent that has access to "tools" (functions) to draw shapes and lines on a coordinate system (X and Y axis).The Workflow: I use a "To-Do List" middleware. The AI lists its objectives first (e.g., "Draw border," "Draw box A") and then executes them step-by-step.The Brain: Since LLMs are bad at math, I gave the agent calculator tools to figure out coordinates.Memory: To manage the context window, I used a summarization agent that compresses the chat history so the AI doesn't forget what it’s doing.
Challenges & Failures During This Project?
The biggest challenge?
Cost. Every time the agent thinks or draws a line, it makes a "tool call," which costs tokens (money). Generating one image might take 15 to 20 calls!. To manage this, I used OpenRouter, which allows access to various models more economically. Also, there isn't a specific Indian solution for these inference costs yet, so we have to rely on US-based platforms.
Did you collaborate with anyone?
To be honest, this was a "One Man Army" show. I didn't partner with any organization. It was a personal thought experiment to test an idea.
Is the solution best for the stated problem?
Is my solution the best? Not yet. The current version is very "bare bones". While the approach of using an agent to draw on a canvas is the best way to solve the editing problem, the user interface needs work. I am working on adding "Undo" and "Redo" features, which are vital for humans but hard for AI.
What sources did you refer to for ‘Online Learning’?
I am a self-taught programmer; I haven't paid a single penny for a course. My best teacher? Official Documentation. Whether it is Python or LangChain, I go straight to the docs. I also believe in the "Dead Internet Theory" - the idea that search results are now flooded with AI-generated slop, making Google less reliable. So, I trust source code and documentation over random blog posts.
How do you keep your ‘Online Reputation’ intact?
I keep a low profile. I don't usually promote my projects aggressively. I post my achievements on LinkedIn, and I maintain a personal website, where I upload my projects and experiments.
Impact of Your Solution So Far & What’s Your Vision for the Future?
For now, "Draw AI" remains a successful experiment. I am not looking to turn it into a product immediately because I don't know if the market is ready. However, the learning impact is huge. I am already using the concepts from this project like the agentic workflow - for a new, unreleased project I am working on.
What Advice You Would Like to Give to Fellow Innovators?
Don't just copy what everyone else is doing. Instead of building another AI browser, build something that changes how the internet works!. Think about impact first. Also, everything can be learned on your own, you don't need expensive courses.
Is there any Mentor, Support or Inspiration you have?
I don't have a specific mentor. Most of my work is personally motivated. I learned everything on my own, and I think that’s an example others can follow. I am my own inspiration.
Best wishes for your endeavours! Good luck for the future, proving India’s Got Intelligence!