I've spent the last 10+ years building native Android applications — shipping features, fixing the hard bugs, and living with the consequences of my own architecture decisions in production. Recently I started learning AI Engineering and Python. Rather than working through isolated tutorials, I chose to build a complete, production-style project to understand the full lifecycle of an AI system, end to end. I'm documenting that journey publicly here on GitHub.

After a decade of building Android apps — and consuming AI APIs like any other backend dependency — I wanted to understand what actually happens on the other side of those calls. So I started learning, from the ground up:

• Python
• Data Engineering
• Machine Learning
• FastAPI
• Explainable AI
• Retrieval-Augmented Generation
• MLOps fundamentals
• GitHub Actions
• CI/CD

The goal isn't to become someone who just calls AI APIs. It's to understand how production AI systems are actually designed, trained, explained, tested, and served — the same way I'd want to understand any system before I ship it. I'm early in this, and I'd rather be upfront about that than overstate it.

Things I'm actively learning right now:

• Python
• AI Engineering
• Machine Learning
• FastAPI
• XGBoost
• SHAP
• RAG
• MLOps
• Production AI systems

Android (10+ years)

Kotlin · Java · Jetpack Compose · MVVM · Coroutines · Flow · Room · Hilt · Retrofit

AI & Backend (learning)

Python · FastAPI · XGBoost · SHAP · RAG · GitHub Actions · Git · Markdown

An end-to-end football analytics platform, and the project I'm using to learn AI engineering by actually building it rather than reading about it. It's a complete system, not a notebook:

• A native Android application (Compose Multiplatform) consuming a real backend
• A FastAPI backend serving predictions and an assistant over REST
• A data engineering pipeline that ingests and validates match data before it's used
• An XGBoost pipeline that predicts match outcomes
• SHAP explainability attached to every single prediction, not just the model as a whole
• A retrieval-augmented assistant that answers questions grounded in the platform's own data, running locally
• CI/CD via GitHub Actions
• Documentation, including architecture decisions, kept alongside the code

Building one complete project taught me more than finishing a dozen disconnected tutorials ever did. Tutorials show you a step. A real project forces you to make the step work with the step before it and the step after it — data quality feeding a model, a model's output feeding an explanation, an explanation feeding a UI. That's where I started to see how software engineering and AI engineering actually depend on each other, rather than being two separate skills.

• LinkedIn: linkedin.com/in/sumukha-aithal-77954858
• Email: sumukh.aithal@gmail.com