I pursued independent game development working entirely as a solo developer, Taking full ownership of the engineering pipeline without relying on external teams or generative AI tools as there were none at the time, I built and published complete titles from scratch, including Santa’s Gift Launcher, Kwebbelcat, and Rodge. Developing Santa’s Gift Launcher required implementing 2D projectile physics, continuous trajectory calculations, and responsive collision resolution tailored for constrained mobile environments. In Kwebbelcat and Rodge, I focused on deterministic state machines for enemy behaviours, custom platformer kinematics, and spatial collision checks. Hand crafting every game loop, memory budget, and mathematical subsystem independently reinforced my programming discipline and systems debugging under real-time execution constraints.

The widening deficit in global healthcare human resources, often exceeding one doctor per 2,500 individuals in suburban regions, consistently overwhelms outpatient clinics with minor, manageable symptoms, delaying critical urgent care until conditions worsen. To solve this bottleneck, I dedicated my final-year bachelor thesis project to building DocBot, an AI-powered clinical information retrieval and symptom triage system. Standard foundation models carry severe hallucination risks in healthcare, ungrounded language models cannot be trusted with clinical triage. To ensure verifiable reliability, I implemented an end-to-end Retrieval-Augmented Generation (RAG) pipeline.

The system enables medical practitioners to index their own verified guidelines and literature directly into a cloud-native vector store. Using LangChain, I segmented clinical literature into 500-character overlapping chunks, mapped them into a continuous 384-dimensional semantic space via the all-MiniLM-L6-v2 transformer, and indexed them in a Pinecone serverless database optimized for cosine similarity. For every user symptom inquiry, the system retrieves the top two relevant clinical passages and injects them into bounded system prompt templates. These guardrails strictly constrain the model to output maximum three-sentence home-care precautions and clear clinical thresholds for hospital visits while rejecting non-medical prompts. Serving the application over a Flask WSGI stack accelerated by Groq LPUs, I achieved an end-to-end round-trip latency of under 1.5 seconds. This practical engineering experience bridging high-throughput vector databases, semantic retrieval, and low-latency inference directly underpins my preparation for advanced graduate research in software systems and artificial intelligence.

I have spent several years exploring systems design and automation through my technical YouTube channel, NaMiature, which has grown to over 215,000 subscribers. On this channel, I design and teach automated mechanisms in Minecraft, such as high efficiency resource farms and redstone logic circuits. Engineering these builds requires working within the fixed constraints of the game’s tick cycle, handling asynchronous item routing, and debugging logic bottlenecks to keep systems running reliably. Producing structured breakdowns for a large audience has sharpened my ability to diagnose architectural issues and communicate technical concepts clearly. These habits directly carry over into software engineering and graduate research.