██████╗ █████╗ ███╗ ███╗██╗████████╗ ██████╗ ██╗ ██╗██████╗ ████████╗ █████╗
██╔══██╗██╔══██╗████╗ ████║██║╚══██╔══╝ ██╔════╝ ██║ ██║██╔══██╗╚══██╔══╝██╔══██╗
██████╔╝███████║██╔████╔██║██║ ██║ ██║ ███╗██║ ██║██████╔╝ ██║ ███████║
██╔══██╗██╔══██║██║╚██╔╝██║██║ ██║ ██║ ██║██║ ██║██╔═══╝ ██║ ██╔══██║
██║ ██║██║ ██║██║ ╚═╝ ██║██║ ██║ ╚██████╔╝╚██████╔╝██║ ██║ ██║ ██║
╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝ ╚═════╝ ╚═════╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝
> whoami
Ramit Gupta — Student
Delhi-NCR, India
> cat focus.txt
ML | DL | Computer Vision | Generative AI | AWS
Hospital_Management_System — End-to-End Hospital ML Platform
patient data ──► feature engineering ──► XGBoost ──► LOS prediction
└──► SHAP explainability
│
FastAPI inference API ◄─── Docker Compose
│
Streamlit dashboard ──► AWS EC2
XGBoost predicts Length of Stay. SHAP makes predictions explainable per-patient. FastAPI serves inference. Streamlit visualizes. Docker Compose ties it all together. Deployed live on AWS EC2.
XGBoost SHAP FastAPI Streamlit Docker Compose AWS EC2 Jupyter
TerraStorm — Cloud-Native Land Change Detection
S3 upload ──► Lambda trigger ──► SageMaker CNN ──► change map ──► Django dashboard
└──► SNS alert
Satellite images in. Deforestation, urban sprawl, illegal encroachment — detected and alerted automatically.
Django CNN (Siamese) AWS SageMaker Lambda S3 SNS API Gateway Docker EC2
IMDB_RNN — Sentiment Analysis · SimpleRNN → Streamlit
raw text ──► tokenize ──► pad sequences ──► Embedding ──► SimpleRNN ──► sigmoid ──► [pos/neg]
Full cycle: preprocessing, training, activation stability analysis (ReLU vs tanh), serialization, live Streamlit app.
TensorFlow Keras Streamlit Jupyter .h5 serialization
Transformer-from-Scratch — GPT Built From Zero
Bigram LM ──► positional encoding ──► self-attention ──► multi-head ──► GPT block ──► generate
No HuggingFace shortcuts. Raw PyTorch, character-level, built block by block. Tokenization explored with tiktoken + SentencePiece.
PyTorch tiktoken SentencePiece Jupyter
Spam_Detection — TF-IDF · Deployed Live
text input ──► TF-IDF vectorizer ──► Naive Bayes ──► spam / ham
│ │
vectorizer.pkl spam_model.pkl ← serialized separately
Vectorizer and model stored independently — production-safe inference pattern. Live on Render ↗
Scikit-Learn TF-IDF Pickle Render
|
Languages ML / DL
|
Computer Vision
Generative AI & NLP
|
Cloud & MLOps
Tools & Frameworks
|
-00FFFF?style=flat-square&logo=darkreader&logoColor=black){kind=icon}
┌─────────────────────────────────────────────────────────┐
│ > currently building : end-to-end ML systems │
│ > exploring : LLM internals · RAG · RLHF │
│ > open to : collabs · contributions │
│ > reach me : ramitgupta17@gmail.com │
│ > linkedin : linkedin.com/in/ramit-gupta23 │
└─────────────────────────────────────────────────────────┘