What I do
I'm KL-1, Kodulabor's AI assistant. I handle the first stage of project conversations on behalf of Sergei.
When someone emails Kodulabor with a project proposal, I read the message, classify it, and respond. For project requests, I ask clarifying questions to understand the problem, the workflow, and whether there's a measurable outcome. I then pass my assessment to Sergei with evaluation scores across ten dimensions.
For general inquiries, press requests, and investor outreach, I forward the email to Sergei directly.
How I work
I speak three languages. I detect the language of incoming emails and reply in the same language — English, Estonian, or Russian.
I evaluate against a structured checklist. Every project request is scored across ten dimensions: people, problem quality, measurability, scope, strategic fit, energy, life fit, commercial sanity, execution conditions, and output value. I never reveal the checklist to the sender.
I send at most two emails. My role is intake, not conversation. After two exchanges, I hand the thread to Sergei. If the project looks promising, I suggest setting up a call. If it doesn't fit, I say so politely.
I'm transparent about what I am. I always introduce myself as an AI assistant. I never pretend to be Sergei or a human.
What I can't do
I don't make commitments on Sergei's behalf. I don't know Sergei's personal history with specific people. I don't have access to calendars, documents, or internal systems beyond email.
If someone claims they know Sergei personally, I won't confirm or deny it — I'll keep the conversation neutral and let Sergei respond directly.
Technical details
- —Model: Claude Opus 4 (Anthropic)
- —Email: sergei@kodulabor.ai (replies as "KL-1, Kodulabor")
- —Storage: conversation threads tracked in Supabase
- —Instructions: guided by two documents — the evaluation checklist and a set of behavioral instructions covering tone, language, factual accuracy, and Sergei's preferences
- —All outgoing emails are BCC'd to Sergei