I'm a Data Scientist & AI/ML Engineer who doesn't stop at model.fit() 💅
My brain runs on Statistics + Data Science — which means I actually understand why the model works before shipping it. wild concept, i know.
I go from raw data → statistical reasoning → ML/DL → GenAI/Agents → actual deployed systems. the whole pipeline. no shortcuts. 🛤️
📦 DATA
↓ "okay what are we even working with"
📊 STATISTICS
↓ "let's not fool ourselves with bad assumptions"
🤖 MACHINE LEARNING
↓ "baseline first, always"
🧠 DEEP LEARNING
↓ "now we're cooking"
✨ GENERATIVE / AGENTIC AI
↓ "make it think for itself"
🚀 ENGINEERED SYSTEMS
"ship it or it didn't happen"
the real flex isn't accuracy — it's building something that still works on monday 😤
|
|
|
|
|
End-to-end analytics platform for churn prediction, customer segmentation, lifetime value analysis and retention intelligence 📊 Includes statistical testing, SQL analytics, multiple ML models, SHAP explainability, K-Means segmentation, DVC pipelines and an interactive dashboard.
|
Stateful conversational AI built around a ReAct-style agent architecture with dynamic tool calling, persistent conversation state and local LLM inference 🧩 Includes SQLite checkpointing, multi-thread conversations, token streaming and 4-bit quantized Qwen inference.
|
|
AI-powered document workflow system covering OCR, classification, information extraction, validation, routing and operational analytics 🏢 Uses document ingestion pipelines, OCR engines, DistilBERT / TF-IDF classification, business-rule validation, DVC and Docker.
|
Multi-sensor Earth Observation pipeline for landscape-change detection across Nepal's Bhote Koshi–Trishuli basin 🛰️ Combines Sentinel-1 SAR, Sentinel-2 optical imagery and SRTM terrain features with spatial block cross-validation to reduce geographic leakage.
|
|
Time-series investigation of night-time radiance across seven Indian landscapes using VIIRS satellite observations and geospatial variables 🌃 Uses STL decomposition, change-point detection, econometric modeling, Granger causality and SARIMAX forecasting.
|
Chess engine combining classical search with numerical optimization to dynamically adapt its evaluation strategy 🎮 Uses Minimax, Alpha-Beta pruning, learned feature weights, Conjugate Gradient optimization and Golden Section Search.
|
🔭 the full toolkit (it's a lot fr)
🧠 Deep Learning CNNs · RNNs · LSTM · GRU · ConvLSTM · U-Net · Transformers · Transfer Learning
👁️ Computer Vision OpenCV · MONAI · MediaPipe · Image Segmentation
✨ Generative AI RAG · Embeddings · BERT · LoRA · Local LLMs · Quantization · Tool Calling
📊 Data & Statistics Probability · Hypothesis Testing · Time Series · Statistical Modeling · Feature Engineering · Experimentation
⚙️ Engineering FastAPI · Streamlit · Docker · GitHub Actions · REST APIs · DVC · MLflow
🏗️ Development / Infra React · TypeScript · Tailwind CSS · MongoDB · PostgreSQL · AWS · GCP · Azure · Vercel · Render
🎯 Problem
│
▼
📦 Data ───────► assumptions / quality / leakage
│ "trust nothing, validate everything"
▼
🤖 Model ──────► baselines / features / learning
│ "simple first, always"
▼
📊 Evaluation ─► metrics / validation / failure modes
│ "if you can't explain it, you don't get it"
▼
⚙️ Engineering ► reproducibility / APIs / deployment
│ "notebook ≠ production"
▼
🌍 Real Use ───► does the system still make sense?
"the real test is monday morning"
"what did it learn, why does it work, where does it fail, and can we actually build something useful around it?" — me, every single project 🤔
Research is one part of my technical foundation — particularly in deep learning, computer vision and spatiotemporal modeling 🧬
📝 Spatiotemporal Deep Learning for Pharmacokinetic Mapping from DCE-MRI ConvLSTM-based modeling with temporal feature aggregation for pharmacokinetic parameter prediction.
🏛️ ISMRM Indian Chapter 2026
- Comparative Analysis of U-Net Architectures for Head and Neck Tumor Segmentation Using Cross-Domain Transfer Learning
- Cross-Species Transfer Learning for Brain Tumor Segmentation: From Humans to Mice
✍️ Technical Writing When Models Lie: The Silent Assumptions Behind Accuracy in Data Science
