If you’ve ever spent hours participating in competitive programming challenges or working through cybersecurity learning tracks, you know the frustration. You submit a snippet of code, it fails a security check, and you’re left staring at an error message.
Current AI tools are great at instantly generating the fixed code for you. But there’s a massive problem with that: you don't actually learn anything.
That’s why I am building Achintisalive as part of my Gen AI Academy APAC project.
*What is Achintisalive? *🤖
Achintisalive is an AI-powered educational co-pilot designed to solve the lack of personalized feedback in secure coding.
Instead of just acting as an automated code-fixer, Achintisalive acts as a Socratic tutor. When you paste code with vulnerabilities (like SQL injections, XSS, or logic flaws), it doesn't give you the answer. Instead, it asks guiding questions and provides targeted hints to help you discover and fix the flaw yourself.
✨ Key Features Planned
The Vulnerability Spotter: Paste your code, and the AI flags common security risks and algorithmic inefficiencies.
Socratic Tutor Mode: A step-by-step hint system that gently guides you toward the solution, enforcing active problem-solving.
Interactive Code Remediation: Once you identify the issue, the app validates your updated code and explains exactly why your new approach is secure.
The Tech Stack 🛠️
To keep development fast and the architecture simple, I'm building this entirely with Python and Google Cloud Generative AI tools:
Frontend UI: Streamlit. It allows me to build a clean, interactive chat interface in pure Python without getting bogged down in HTML/CSS/React.
Core AI Engine: Google Gemini 1.5 Flash API (via Google AI Studio). I chose this model because of its incredibly low latency and massive context window—perfect for parsing code quickly.
Deployment: Google Cloud Run. The app will be containerized using Docker and deployed serverless for easy scaling.
**The Biggest Challenge: **Making the AI "Less Helpful" 🧠
The biggest technical hurdle I'm tackling right now is prompt engineering. Out of the box, generative AI models want to be too helpful. If you show them broken code, they immediately rewrite it for you.
To solve this, I am engineering strict System Instructions within Google AI Studio. I have to design a robust prompt that completely restricts the Gemini model from outputting raw code fixes, mandating that it must respond with educational hints and guiding questions.
Build in Public: Work in Progress 🚧
I am building this project in public right now! I haven't written the final lines of code yet, but the repository is set up and the architecture is mapped out.
If you are interested in Gen AI, secure coding, or educational tech, I’d love for you to follow along.
🔗 Check out the repo here: achinttiwari/achintisalive
(Feel free to drop a ⭐ if you like the concept!)
