Why my landing page is an agent, not a résumé
I replaced the usual hero-and-bio layout with a bounded conversational card. Here's what it does, what it cost, and what I'm watching for.
The usual personal site asks a visitor to read about me. This one asks them to talk — the landing page is a small conversational card, grounded in curated notes about my projects and how I work, that tries to find an overlap with whatever the visitor brings.
A few decisions that shaped it:
- Bounded on purpose. Ten responses per session, a monthly request budget enforced server-side in Redis, and a soft handoff after three responses that points to me, the human. The card's job is to start a conversation, not to be one.
- Grounded, not improvised. The agent only knows what's in a set of markdown files I review — profile, projects, principles, boundaries. When it doesn't know, it's instructed to say so and point at me.
- Honest fallback. Without a model available it degrades to a read-only mode with pre-written answers, and the static build log is always there for people who would rather read than chat.
The interesting product question: does a 60-second exchange with an agent produce a more specific first conversation between two humans than a CV skim does? That's the bet. I'm taking it to Product at Heart in Hamburg to find out.