Proof point → I turned a marketing exercise - “know your ideal customer” - into a living AI agent I can interrogate. That’s how study becomes infrastructure.

🩻 Problem

Founders are told to “know their ideal customer,” and the knowledge dies in a slide. Meanwhile, agent frameworks make it possible to embody that customer - if you put the creative effort where it matters.

🔨 Solution

A personal AI agent system standing on the elizaOS framework:

Architecture Overview

  1. Framework, not from scratch - @elizaos/eliza-starter with the full plugin suite: Postgres/SQLite adapters, Discord/Telegram/Twitter/direct clients, image generation, Tavily search, and AMQP - deployed with Docker, docker-compose, and pm2 service scripts.
  2. The persona layer is the product - hand-authored character files. The standout is Fred: a mid-30s European startup shareholder with $7,500 in savings, mid-fundraise, afraid of missing product-market fit - running live on Discord via OpenRouter. Fred is the ideal-customer profile from “SECRET-08: Meet F.R.E.D.” in my startup-copy study notes, made conversational.
  3. Multi-client, model-routed - one persona, consistent across chat surfaces, with model selection handled through OpenRouter.

Honest scope: a one-sitting build (March 2025) on an open framework - the engineering judgment is in what not to build.

📜 Philosophy

Stand on open infrastructure; spend creativity on the differentiated layer. And close the loop between learning and building - book notes in git become agents in production.

🎓 Key learnings

  • Configuring multi-client LLM agents and model routing.
  • Character engineering: encoding bio, goals, and fears so an agent behaves consistently in character.
  • Dockerized operation of long-running agent processes.

📈 Output & impact

  • A working, deployable agent system and a reusable method: turn any customer-research artifact into an interviewable persona.

🌍 Why this matters

Original Tools & IP. The cheapest scalable mentor, support rep, or research subject is a well-built agent persona. The method here - customer research → character file → multi-channel deployment - transfers to onboarding coaches, product explainers in local languages, and user-research stand-ins.

🚀 Hire me

Want a persona-driven agent your team can actually interrogate? Let’s talk → · See also: Ketu Wok · The thesis