I'm applying for the Literature Specialist — Freelance AI Trainer project with Meridial Marketplace, by Invisible. The brief is one I can demonstrate rather than describe — challenge advanced language models on literary reasoning, document where they fail, and give structured feedback that improves interpretive accuracy — so rather than just describe how I'd approach it, I built a live application hub that starts doing exactly that. You can review it before we speak:
Let me be straight about who I am, because the standard I work to starts with honesty. I'm an English Literature graduate and a lifelong close reader — and, genuinely, a practising AI trainer who documents model failure modes every single day. That second half isn't a stretch for this role; it's my actual work. I run a multi-model workflow (Claude, Gemini, Grok) and my whole method is built on catching confident-but-wrong output, writing reproducible error traces, and feeding back the corrected reasoning. Door 02 of the hub is a live literary red-team engine — adversarial prompts across six literary domains with the models' failure modes documented; Door 03 is the taxonomy and scoring rubric I'd grade those failures against.
What makes me a credible fit on the literature side specifically: an English Literature degree gives me the vocabulary and the habits — theme, narrative structure, symbolism, genre, close reading — and years of evidence-first analytical work (a Law degree, then Australian government service) trained me never to assert what I can't ground in the text. The metacognitive, "show your work" communication the brief asks for is simply how I already write — every trace narrates the interpretive steps, not just the verdict.
Where I'd be honest about my limits: I'm not a tenured literary-theory academic or a published critic. At the frontier of formal critical theory, or in deep specialist period and comparative scholarship, I read as a strong informed generalist and would flag for a specialist rather than bluff — which is exactly the failure mode this role exists to catch.
I'm available immediately, fully remote, set up with secure compute and high-speed internet, and comfortable working to evaluation guidelines and calibration at pace. I'd welcome a short call — or, better, a paid trial task, since the fastest way to judge a trainer is to read a real trace.