Sahaj
Gyawali

Performed exploratory data analysis (EDA) on COVID-19 datasets to identify global and country-level trends.

Visualized confirmed, deaths, recovered, and active cases worldwide and specifically for Nepal using line plots and bar charts.

Created heatmaps to compare countries and WHO regions, highlighting concentrations of confirmed cases, recovery rates, and active cases.

Implemented sorting functions to identify the top 5 countries by confirmed, recovered, active cases, and deaths.

Focused on clear, insightful visualizations using Matplotlib and Seaborn for actionable interpretation of pandemic trends.

Created prediction model using RandomForestRegressor to predict house price with high accuracy.

Performed data preprocessing: handling missing values, encoding categorical features, and feature scaling.

Utilized Pandas and Numpy for manipulation; Matplotlib & Seaborn for feature visualization.

Implemented Scikit-learn for model training, hyperparameter tuning, and performance evaluation.

Developed a full-stack recommendation engine using content-based filtering with Cosine Similarity logic.

Engineered a custom web scraper using BeautifulSoup to aggregate movie metadata and plot synopses.

Built an interactive UI using Streamlit to provide real-time recommendations and live movie details.

Implemented NLP techniques for text vectorization using Scikit-learn to measure feature similarities.

Developed a sophisticated LLM-powered content tool that bridges raw data analysis and creative LinkedIn post generation.

Implemented a 'Data-to-Draft' pipeline using preprocess.py for metadata extraction, hashtag unification, and style alignment.

Built a few-shot learning engine (few_shot.py) to retrieve contextually relevant examples from processed_posts.json for dynamic prompt engineering.

Integrated LangChain with Groq's llama-3.3-70b-versatile model for fast, style-aligned content generation.

Delivered an intuitive Streamlit dashboard allowing toggling of post length, language (English/Nepali), and specific tags.

Strategic enhancements planned include Engagement Prediction (Viral Score), Semantic Style Matching with vector embeddings, and Automated Image/Graphic pairing for complete content automation.

Developed a production-ready churn prediction system bridging research notebooks to an interactive UI.

Implemented a clean separation of concerns: Pydantic for data validation, Model Service for business logic, and FastAPI + Streamlit for delivery.

Loaded and served trained ANN model (.keras) and scaler (.pkl) using a Singleton pattern for efficient inference.

Enhanced UX with churn probability, risk levels, and retention recommendations displayed on a real-time Streamlit dashboard.

Strategic enhancements include planned integration of SHAP/LIME for model explainability and GeoJSON-based province-level churn visualization.