kevin officialz

I’m an aspiring Machine Learning Engineer who builds domain-focused AI with an emphasis on real-world problem solving—especially NLP and data-driven modeling. I’ve been developing hands-on projects end-to-end, from data preparation and model training to evaluation and iteration.

In Feb 2026, I fine-tuned a LLaMA-based Large Language Model on a 20K-sample legal dataset (court judgments & statutory Q&A) using LoRa (PEFT) with Unsloth. I built an instruction-tuning pipeline, optimized LoRA training using 8-bit AdamW and mixed precision for efficient GPU utilization, and evaluated performance with legal reasoning test cases (loss convergence ~1.82 final loss).

Before that, I developed an unsupervised ML system for dynamic AQI clustering (Nov 2025), adapting to seasonal variations (Summer, Winter, Monsoon) rather than static clustering. I implemented season-wise K-Means with standardization and the Elbow Method, and used EDA with pollutant contribution analysis to uncover dominant pollutants and seasonal pollution patterns.

I also enjoy building user-facing solutions: for SIH 2025, I designed AyurSutra—an Ayurvedic Panchakarma patient management platform with a doctor–patient interaction system to streamline the full care workflow. I complemented my development skills with a 4-week UI/UX training (Figma, wireframing, prototyping, usability testing), creating an EV Charging Station Finder & Booking App.