Soaeb Hasan soaebhasan12

Computer Science Student | Python Developer | AI/ML Enthusiast 🇮🇳

💫 About Me

3rd year B.Tech. CSE student building full-stack and ML projects. Experienced in Python, Django, React, React-Native, Node, Express, and Machine Learning. Open source contributor (Kolibri). Mentor at GSSOC. Looking for backend/ML engineering internship where I can solve real problems and learn from experienced teams.

💻 Check out my portfolio:
- 🌐 SOAEB HASAN
📫 Reach me at: hasan.soaeb.ali@gmail.com

💼 Experience

Open Source Contributor | Kolibri (Learning Equality) (Jun'26 - Present)
- Migrating and testing Vue component test suite from @vue/test-utils to Vue Testing Library
- Designed 8+ comprehensive test cases covering user interactions, async operations, and error handling
- All tests passing with code review approval from maintainers
- Learning: VTL best practices, test-driven development, Git workflow
Android Malware Detection using ML | Internship IIT Roorkee (Feb’26 - May’26)
- Developed multi-view transformer model (MalBERT-XAI) for Android malware detection using PyTorch
- Achieved 99.5% binary accuracy and 94.3% family classification across 5 malware families
- Built full ML pipeline: APK parsing → feature extraction (permissions, API calls, opcodes) → tokenization → model training → evaluation
- Implemented interpretability layer using LIME + SHAP to explain model predictions
- Key Learning: Transformer architectures, multi-task learning, model interpretability

🚀 Projects

Android Malware Detection using Multi-View Transformer (Python | PyTorch | Transformers | Deep Learning | NLP | DistilBERT | LIME | SHAP)
- Built MalBERT-XAI, a multi-view transformer architecture with cross-attention fusion for Android malware detection, achieving 99.5% binary accuracy and 94.3% family classification
- Engineered a 4-view feature pipeline extracting permissions, API calls, intents, and opcodes from APKs, replacing single-input approach and improving accuracy from 91.6% to 99.5%
- Implemented a 3-level explainability framework (attention weights + SHAP + LIME) to interpret model decisions at view-level, global, and token-level granularity
AI-Powered Organ Matching Platform (Django | Python | Scikit-learn | Tailwind | HTMX | Alpine.js) - Live Project
- Problem: Manual donor-recipient matching is slow and error-prone in organ transplant systems.
- Solution Built:
  - Designed ML pipeline matching donors to recipients using TF-IDF + cosine similarity on 15+ clinical features (blood type, urgency, proximity, etc.) — 95% accuracy on 900+ records
  - Full-stack: Backend REST API (Django) + Frontend (Tailwind/HTMX) + ML model serialization
  - Implemented business rules layer (blood type constraints, urgency prioritization) on top of ML scores
  - Achieved sub-second inference time for real-time recommendations in production environment
- Key Learning: Balancing ML accuracy with domain-specific business rules
Boston Housing Price Prediction Model (Pandas | Numpy | Scikit-learn | Matplotlib)
- Problem: Given historical house data, predict future housing prices with confidence.
- Solution Built:
  - Exploratory Data Analysis: Identified 5 key price-influencing features (location, age, crime rate) using correlation analysis and statistical tests
  - Feature Engineering: Created polynomial features and handled outliers — improved baseline by 12%
  - Model Development: Trained multiple regression models (Linear Regression, Ridge, Lasso),
    selected best using cross-validation (R² = 0.85)
  - Visualization: Created heatmaps and scatter plots to communicate findings
- Key Learning: ML project workflow — EDA → feature engineering → model selection → validation
BarberBook — Full-Stack Mobile Booking Marketplace (React Native | Expo | Node.js | Express | MongoDB | Razorpay | Firebase FCM) - GitHub Repo
- Problem: Local barbershops and salons rely on phone calls and walk-ins for bookings, leading to double-bookings, no-shows, and no centralized way for owners to manage staff, services, or daily schedules.
- Solution Built:
  - Built a two-sided marketplace app end-to-end (mobile frontend + REST API backend) supporting separate customer and shop-owner experiences on one codebase
  - Implemented phone-number + OTP authentication with stateless JWT sessions, and Razorpay payment integration with server-side HMAC-SHA256 signature verification
  - Designed a real-time booking engine with timezone-safe (IST) slot-availability logic and Firebase Cloud Messaging push notifications, including per-device token scoping to prevent cross-account notification leaks
  - Executed a full app-wide UI/UX redesign (color system, iconography, navigation) through a single centralized design-token theme file, avoiding hardcoded styling across 10+ screens
- Key Learning: Designing for statelessness (JWT) and single-source-of-truth theming early made both security and large-scale visual changes dramatically cheaper later in the project.