Thomas A.

With a background in Applied Linguistics and 5 years of educational experience, I specialise in the fine-tuning of Large Language Models (LLMs) to ensure they move beyond robotic accuracy toward genuine Emotional Intelligence (EI). My approach: Trauma-Informed & Safe: I design Safety Frameworks and "Graceful Decline" (Punt) protocols for sensitive YMYL (Your Money/Your Life) topics, ensuring users feel validated without receiving unauthorised advice. Cognitive Load Focused: I prioritize Digestibility over simple conciseness. I train models to provide structured, scannable information that reduces the learner's mental effort. The Humble Interpreter: I specialize in Contextual Disambiguation, forcing models to ask clarifying questions when user intent is "borderline" rather than hallucinating an interpretation My principles: I provide the high-fidelity signals needed for Supervised Fine-Tuning (SFT) and RLHF (Reinforcement Learning through Human Feedback), specialising in the creation of "Golden Responses" that serve as the gold standard for model behaviour. I design prompts to avoid "Slog" behaviors (tedious, low-value tasks like massive list sorting) in favor of Advanced Reasoning and Multi-source Synthesis. I also focus on moving model outputs from 'Okay' (adequate with minor flaws) to 'Amazing' (flawless, natural, and accurately following all implicit and explicit instructions).