From 2023–2025, Lee worked as an AI Content Engineer at Meta within the firm’s GenAI team/Superintelligence Labs, supporting the development and fine-tuning of generative AI outputs for the Llama model family. The role combined editorial judgement with product thinking—testing AI model behaviour, improving accuracy and tone, and translating what worked into repeatable systems.

Work included:

  • Claim extraction and verification: identifying the core factual claim in a model response, verifying it against reliable sources, and scoring truthfulness to improve training data and reduce hallucinations.

  • Meta AI conversation optimisation: writing and iterating welcome messages, icebreakers and null-state prompts to improve onboarding, engagement and retention.

  • Tone and QA: running quality checks on accuracy, clarity, empathy and safety; flagging failure modes (tone drift, preachiness, repetition, unsafe outputs) and working with product and engineering teams to implement improvements.

  • Golden Set development: creating “best practice” prompt/response exemplars and tone guides—rewriting everyday responses into the ideal Meta voice so the model had consistent references to learn from.

  • AI Studio character creation: building public-figure and fictional personas using structured prompts, backstories and dialogue styles; ensuring outputs remained persona-true while staying within product, safety and IP constraints.

  • Group chat activities/scenes: prototyping multi-character activities (games, roleplay narratives, advice panels) and refining flow and pacing through testing and iteration.

  • Operationalising quality: packaging learnings into templates, checklists and reusable prompt structures to help teams scale consistent outputs across products.

Bringing quality control to GenAI content

With hands-on experience shaping GenAI outputs at scale, Lee brings an editorial approach to making AI-generated content clearer, safer and more authentic. The same principles that guided his work at Meta — accuracy, tone, user intent and quality control — can be applied to helping brands and agencies use GenAI without compromising credibility or brand voice.

Get in touch if you’d like support with voice and tone, AI content QA, prompt frameworks, or the practical know-how to keep GenAI content on-brand, authentic feeling and press-ready.