I'm an ML developer from India who builds end-to-end AI systems — not just models, but full working applications with backends, UIs, and real results.

I completed a 9-month AI minor from IIT Ropar covering ML, deep learning, NLP, and computer vision — and I've been building independently ever since.

• 🔭 Currently building a Chatbot with Rasa and a Book Recommender System
• 🌱 Learning deployment and MLOps basics
• 💬 Ask me about Python, Scikit-learn, NLP, Flask, XGBoost
• 📫 Reach me at kskirtisharma2000@gmail.com
• ⚡ Looking for internships & fresher roles in ML/AI

Classifies messages as spam or ham with 97% accuracy using an ensemble Voting Classifier

• Full NLP pipeline: tokenization, stopword removal, stemming, TF-IDF vectorization
• Combined SVM + Decision Tree + Extra Trees into a Voting Classifier
• Deployed as a live interactive web app using Streamlit

Predicts emotional state from journal text + metadata and recommends personalised actions

• Combined TF-IDF (5000 features, bigrams) with metadata via sparse matrix stacking
• Used Logistic Regression (tuned with GridSearchCV) for emotion classification
• Used XGBRegressor (tuned with RandomizedSearchCV) for intensity prediction (scale 1–5)
• Built a rule-based decision engine — maps state + intensity + context to recommended actions with timing
• Added uncertainty modeling — flags predictions with confidence below 0.55
• Ran 3-way ablation study: combined text + metadata outperformed both text-only and metadata-only baselines
• Deployed with Flask backend and HTML/CSS frontend

Conversational AI chatbot with NLU pipeline, Flask backend, and custom frontend (In Progress)

Two-engine recommendation system — popularity-based + collaborative filtering on 1.1M+ real ratings

• Processed 1.1M+ ratings across 271,360 books and 278,858 users (Amazon Book-Crossing dataset)
• Built user-item pivot table (706 books × 810 active users); applied cosine similarity to generate top-5 personalized recommendations
• Input 1984 → returns Animal Farm, The Handmaid's Tale, Brave New World — based on real user rating patterns
• Popularity engine surfaces top 50 books filtered by minimum 250 ratings and average score; deployed via Flask + HTML/CSS

I'm open to internships, fresher ML roles, freelance AI projects, or just a good conversation about AI.