Sofia Fatima
Hyderabad, India | +91 8519995787 | sofiaafatima20@gmail.com
Professional Summary
Highly motivated, enthusiastic, and analytical Computer Science student and Java Developer (8.59 CGPA) bringing creativity and problem-solving expertise to backend architecture, data processing, and building robust SaaS applications. Demonstrated success in delivering automated AI/ML solutions, ensuring data protection, and troubleshooting complex systems.
Projects & Development
Nano-Mirror AI (ARC-AGI Solver)
Apr 2026
Tech Stack: PyTorch, ONNX, JavaScript, GitHub Pages
- Engineered a highly optimized Convolutional Neural Network (CNN) with sub-1,000 parameters (<50KB footprint) to solve ARC-AGI visual reasoning tasks with 100% accuracy.
- Deployed a zero-dependency web application utilizing ONNX Runtime for real-time, browser-based edge inference without relying on cloud computation.
- Resolved spatial boundary limitations by designing custom depthwise and pointwise convolutions, ensuring mathematical purity in geometric mapping.
Vidyut – AI Gamified Learning Platform
Apr 2026
Tech Stack: Flutter, Firebase, Gemini AI API
- Designed and deployed a cross-platform educational application (Native APK & Live PWA) integrating Large Language Models (LLMs) including the Gemini AI engine.
- Engineered a gamified freemium user model with secure authentication, optimized state management, increasing user retention by 30%.
- Implemented production-level deployment practices by securing environment variables against repository scrapers.
Enterprise FinTech Backend Architecture
Mar 2026 – Apr 2026
Tech Stack: Node.js, Express, Sequelize ORM, PostgreSQL
- Architected a production-ready financial API handling up to 500+ concurrent requests, ensuring Data Integrity and ACID compliance.
- Implemented Role-Based Access Control (RBAC) and data aggregation logic to provide real-time financial insights securely.
Network Intrusion Detection System (NIDS)
Jan 2026 – Mar 2026
Tech Stack: Python, Scikit-Learn, Pandas
- Developed a predictive model using the NSL-KDD dataset to classify network anomalies with 98% accuracy.
- Processed real-time network packets with sub-second latency (0.8s), demonstrating proficiency in large-scale data processing and cleaning.