.
Submission for the Hacktoberfest Open-Source AI Challenge — Week 2: Side Hustle
🚀 What I Built
Ruko is a privacy-first AI assistant that helps people identify suspicious SMS messages, understand scam warning signs, and make safer decisions. The core safety checker is free, with potential paid family and small-business setup services.
🎬 Demo
https://drive.google.com/drive/folders/1h7UnWrpRNzhmHBeUbQY-W3ahjW_VG_Ix?usp=drive_link
💻 Code
GitHub: https://github.com/vedant21-ctr/rukoAI
⚙️ How I Built It
Built with Python, FastAPI, Next.js, TypeScript, Streamlit, Ollama, and Gemma 2B.
Ruko combines a privacy gate, safe structured signals, deterministic risk scoring, local AI explanations, and a safety validator. It also includes an interactive educational lab and a development-only subscription simulation.
🌍 Why Open Innovation Matters
Local inference helps keep message content on the user's device and reduces dependence on paid cloud AI APIs. Open-weight models also give developers more control over model selection and experimentation.
👤 My First User
Ruko is ready for demonstration, but real-user feedback and willingness to pay still need validation. No paying customers are claimed yet.
🏆 Prize Categories
- Gemma: Local Gemma 2B integration via Ollama.
- Overall Side Hustle Challenge: Privacy-first AI with a potential commercial model.
📊 Evaluation
The current hybrid benchmark achieved 82.54% recall and 61.86% accuracy on 118 classified samples. False positives remain a major limitation, and improving detection quality is a priority.
The project reports 71 passing tests and a successful Next.js production build.
Building Ruko has reinforced one principle: AI security tools need privacy, transparent evaluation, and safe fallbacks—not just a model that produces convincing answers.
Stop. Check before you tap. 🛡️
Top comments (0)