Physicist turned oceanographer turned data scientist — I bring a scientific instinct for patterns and a researcher's rigour to ML and AI.
I build end-to-end intelligent systems: from raw data to deployed models, from EDA to Streamlit apps, combining computer vision, NLP, and generative AI to solve real-world problems.
📍 Kochi, Kerala | 🔗 LinkedIn | 💻 GitHub
BSc Physics (9.24 CGPA) → MSc Physical Oceanography, CUSAT (8.29 CGPA)
→ Project Associate, DRDO-NPOL (Python + MATLAB on large-scale satellite data)
→ Data Science & AI (now)
An unconventional path — and a deliberate one. Scientific computing at DRDO gave me hands-on experience with real, messy, large-scale datasets long before I wrote my first ML model. That foundation shapes how I approach every data problem.
- MedAI Nexus — multi-modal AI healthcare platform (CV + ML + OCR + LLM)
- AI automation workflows and agentic AI systems
- RAG-based applications and AI-powered analytics tools
- Agentic AI & autonomous workflow systems
- RAG (Retrieval-Augmented Generation) pipelines
- AI app & product development
- Advanced generative AI with real-world deployment
Languages & Databases
ML / Data Science
AI & Deep Learning
Visualization & Analytics
Tools & Platforms
Multi-modal AI healthcare platform integrating Computer Vision, ML, OCR, and LLM-based assistance
Tech: Python · TensorFlow · XGBoost · Tesseract OCR · Gemini 2.5 Flash · Streamlit
| Module | Approach | Result |
|---|---|---|
| Skin disease classification | MobileNetV2 transfer learning, 20 classes | Resolved 6.6× class imbalance with Focal Loss |
| Diabetes risk prediction | XGBoost on 253K CDC health records | AUC-ROC: 0.803 |
| Medical report analysis | Tesseract OCR + multi-step preprocessing | OCR accuracy: 40% → 85% |
| AI health chatbot | Gemini LLM + personalised recommendations | Deployed on Streamlit Cloud |
ML solution for automated agricultural sorting using morphological features
Tech: Python · Scikit-learn · Random Forest · SVM · KNN · CatBoost · Jupyter
- Trained and benchmarked 5 classification algorithms on area, perimeter, shape, and texture features
- Achieved 89% classification accuracy on multi-class dry bean data
Python CLI application with authentication, CRUD operations, file handling, and validation
| Certification | Issuer | Year |
|---|---|---|
| Data Science with Generative AI | Illinois Tech, US | 2025–Present |
| Data Science with Generative AI | Entri, NSDC | 2025–Present |
| Meta Data Analyst Specialization | Coursera · Meta | 2026 |
"Turning data into decisions — one model at a time."