Ask the site.
A language model that knows what's on this site — background, work, education, projects — sitting one floor below the page, ready to answer questions. Each message is routed through OpenRouter's free-tier router, which picks a random free model per request; the chat reports back which one actually answered.
01The models
The chat targets openrouter/free, a meta-router on OpenRouter that randomly assigns each request to one of its free-tier models (Llama, Gemini, DeepSeek, Qwen, and others). Every reply is tagged with the actual model that produced it, so the variation is visible. Requests go through a small Next.js edge route, so the API key never reaches the browser.
02The context
Everything in content.ts — career, education, projects, publications, the quote — is built into a structured persona prompt and handed to the model as its system message. No retrieval, no embeddings. The whole biography fits comfortably in the context window, so the model just has it.
03What it knows, and doesn't
It can answer questions about roles, dates, what was shipped, the shape of the publications, where Lymperis studied, the languages and stacks he's worked with, and his general approach to engineering and leadership. If the question isn't in its notes, it'll say so rather than make something up. It also won't pretend to be Lymperis — it's an assistant about him, not him.
These are free-tier models, so expect occasional rate limits or the odd awkward sentence. For anything serious — collaborations, roles, conversations — write to lymperis.perakis@gmail.com directly.