Final year B.Tech. CSE student building full-stack and ML projects. Experienced in Python, Django, React, Machine Learning and Pytorch. Open source contributor (Kolibri). Mentor at GSSOC. Looking for backend/ML engineering teams where I can solve real problems and learn from experienced collaborators.
- 💻 Check Out My Portfolio v1.0 — Evolution in progress:
- 📫 Reach me at: [email protected]
-
Machine Learning - Python | 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
-
Open Source Contributor | Kolibri (Learning Equality)
- 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
-
GovVault: Local-First RAG for Government Documents | (Django, React, pgvector, Celery) Repo | Live
- Built a RAG assistant that parses PDFs (PyMuPDF), generates on-device embeddings (SentenceTransformers) and retrieves top matches from pgvector
- Returns page-level cited answers via Groq; async ingestion with Celery/Redis
- Deployed as a Django REST API with React/Tailwind frontend
-
Stereo Distance Estimation with OpenCV | (C++/Qt, OpenCV) Repo
- Developed a desktop application for stereo-camera calibration and rough distance estimation using OpenCV and Qt
- Implemented camera calibration workflow with chessboard patterns, intrinsic/extrinsic parameter estimation, and image rectification for stereo alignment.
- Gained practical exposure to epipolar geometry, triangulation, and 3D perception fundamentals relevant to computer vision research.
-
Malware Detection - Machine Learning (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
-
Pothole Detection and Depth Estimation System | (Python, OpenCV, YOLO, SQLite) Repo
- Developed a computer vision pipeline to process video and live camera feeds for pothole detection using YOLO-based object detection.
- Integrated MiDaS for relative depth estimation to assess pothole severity and improve scene understanding in road environments.
- Built an end-to-end system for detection tracking, timestamped logging, and storage of results in SQLite for analysis and reporting.
- Worked on real-time perception, vision-based anomaly detection, and practical deployment of deep learning models in a real-world driving context.
-
AI-Powered Organ Matching Platform (Django | Python | Scikit-learn | Tailwind | HTMX | Alpine.js)
-
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
-
|
|
|
if socials == "Connected":
|
|
|
| 🌐 Web Development | |||||
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
|
| 📱 Software Development | |||||
 |
 |
 |
 |
 |
|
| 🧠 Machine Learning & Cybersecurity | |||||
 |
 |
 |
 |
 |
|
| ⚙️ Backend & Frameworks | |||||
 |
 |
 |
 |
||
| 🗄️ Databases & Tools | |||||
 |
|  |
 |
 |
|
| 🧠 DevOps | |||||
|
 |
 |
|||
|
|
|
|
|
|
|
|
|
|
|
|
|
