Hi, I'm Yashu 👋
A cloud engineer learning to build and showcase my work.
A cloud engineer learning to build and showcase my work.
Master's student in Data Science with 4 years' experience as a Cloud Engineer and Implementation Consultant at ArisGlobal, cutting deployment time and project go-live cost by 15%. Strong hands-on foundation in SQL and Power BI — writing queries to extract, join and aggregate data from relational databases, and building dashboards and data models that turn raw operational data into clear, actionable insights. Experienced partnering with technical and non-technical stakeholders across Finance and Operations to document data logic and surface process inefficiencies. Now looking to bring that analytical rigour to data-driven roles across analytics, machine learning and business intelligence, with a strong interest in e-Mobility and fast-moving product environments.
A data-visualization study of global maritime trade (2019–2026) using IMF PortWatch's daily, satellite-derived AIS data for ~90,000 vessels. Answers 10 analytical questions on how the 2023–24 Red Sea and Panama Canal disruptions rerouted trade rather than shrinking it, via a shared pipeline feeding both a notebook and a live dashboard.
A hybrid caching layer for LLM applications that checks an exact-match cache, then a semantic cache (sentence-embedding similarity via FAISS) before falling back to a local LLM — cutting latency, cost, and repeated inference. Includes a benchmarking and threshold-tuning framework for cache-hit quality.
A machine-learning system that recommends the best-suited crop for a plot of land from its soil nutrients (N/P/K) and climate (temperature, humidity, pH, rainfall), trained on the Kaggle Crop Recommendation dataset (2,200 records, 22 crop classes). Ships as a Streamlit app with explainable feature importances plus a Power BI dashboard for exploring the data.
An end-to-end ML pipeline for weakly-labelled bird-sound classification, built around the Kaggle BirdCLEF 2026 dataset. Converts field-recorded audio into Log-Mel spectrograms and classifies them with a 2D CNN using Multiple-Instance-Learning pooling over frame-level predictions, since only file-level species tags are available. A Streamlit dashboard lets you upload audio and view spectrograms, species predictions, and confidence scores.
A full-stack decision-support system for retail inventory management: a normalized multi-store, multi-supplier PostgreSQL schema with live stock-health monitoring, reorder alerts, and days-to-expiry tracking. Derived-field logic (stock status, reorder alerts, capacity) is implemented once on the backend and imported directly by the frontend so the UI can never disagree with the API.
Recognized for delivering the Roche project go-live with zero production issues.
Achieved the title in the ArisGlobal ELEVATE award for the successful implementation of four live projects within a brief timeframe.
Awarded for rapidly mastering core cloud technologies and contributing to the live production environment within a short onboarding period.
Email: yashubopannapd012@gmail.com