AI Consultant Representative Platform

The platform exists because the way consultancies sell stopped working. AI is changing what clients buy, and how they decide who to buy it from. A static CV plus an uppdragsbeskrivning, the way Swedish consulting has presented people for twenty years, assumes a buyer who has time to read, infer, and translate someone's listed engagements into "can this person solve my problem". That buyer is gone. Listing titles, years and clients does not survive the question the modern buyer actually asks.

Alfija's bet is that the unit of consulting selling is no longer the CV, it is the conversation. Each consultant gets their own AI agent, trained on the projects they have actually delivered and the depth of their expertise, and that agent represents them directly to a client. The interaction stops being a one-way pitch and becomes an intimate discussion. The client asks the question they actually have, and gets a real answer in the consultant's voice, on demand, without scheduling anything first.

Each consultant has an agent. Every agent has a system prompt grounded in that person's actual background, banking and AML for one, transformation methodology and public-sector AI for another, project leadership and solution architecture, automation strategy with RPA depth, fullstack engineering with RAG and Azure. The roster is five today and growing. A separate concierge agent on the landing page knows the team in detail and routes traffic: match the visitor's problem to the right person, explain why, hand off, never list everyone for every question.

The funnel has three doors. From the landing page, a visitor can pick one of three named flows, Discover (we have a problem we can't articulate), Explore (show us how you've solved similar things), Scope (let's sketch a collaboration). Each flow seeds the concierge with a structured opener and the agent recognises the prefix and switches mode: ask diagnostic questions, lead with project markers, or run a scoping protocol. There is also a separate AI Readiness Assessment with a 6-question form that produces a structured analysis, readiness score out of 10, three concrete next steps, the best consultant match, two to three relevant project cards. After three user messages on any chat, a quiet CTA bar appears with three actions: email a summary, request a proposal sketch, or book a call. All three open the same reusable lead modal.

The architecture is deliberately layered. Browser hits nginx, which serves the static frontend and reverse-proxies API calls to a FastAPI backend. FastAPI talks to a Node.js agent proxy, a persistent process that holds long-lived openclaw acp child processes per session over stdio JSON-RPC. Behind that the OpenClaw gateway runs as user systemd, with two ChatGPT Team OAuth profiles round-robined for quota.

In practice. The agent streams from the first token, not after a cold start, second messages come back in seconds, not half-minutes. Project evidence surfaces inline on demand: when a one-line description is not enough, the agent pulls the full project context, challenge, approach, outcome, metrics, team, and answers properly informed, with a project card appearing mid-conversation. Two visitors in two tabs are two independent conversations; nothing bleeds across sessions. And when someone lands directly on a project page, the agent already knows what they are reading and builds on it instead of repeating it.

The platform is not a chatbot. It is what every consulting buyer claims to want and rarely gets, a real conversation with the right person, on demand, in the consultant's actual voice, wrapped in a sales surface that converts attention into a lead with three named exits. The consultancy sells AI transformation. Their own front door is now a working demonstration of it. That is not an accident.