Meeting Notes — 2026-04-23

Participants: Max, Reiner, Maha Duration: ~62 min Video: YouTube

Context

Reiner + Maha about to leave on a longer trip — several weeks of alumni gatherings and networking across Europe. Between now and mid-May they'll be mostly travelling, but meeting lots of people.

Since the Apr-16 call they met two new contacts who independently pointed at our direction:

• A business connector (former co-founder of an early entrepreneurship network, introduced via a mutual contact). Sits on several boards, currently heavy on AI. When Maha described the Zurich setup (AI intervening in the quality of the conversation, not just note-taking), he immediately wanted a demo for a client of one of his boards. His rationale: today's note-takers train the AI but don't serve the people in the room — he wants that flipped.
• A trained coach + tech developer — building a ChatGPT-based coach; has mapped 17 coaching modalities and wants the agent to detect which modality fits the client's situation. When we explained we work on human-to-human conversations with a psychological model attached, he saw the frame: "you're building a PlayStation — the games (psychological models) plug in." Shared our one-pager with his co-founder on the spot.

Both want to stay in partnering mode.

Demo 1 — "Smart circle" conference format

Max walked through a registration flow for a non-conference where AI forms conversation circles instead of scheduling talks.

Registration captures

• Who you represent, topics of interest, primary intent (mentor / learn / collaborate), who you want to meet, what you bring, what you want to receive.
• Diversity preference (deep dive with similar people vs. cross-sector; seniority weighting).
• Session count and accessible languages.

Output

• Circles start forming based on matches + diversity criteria; at ~5 registrants a timing slot opens and the circle convenes.
• Every circle should produce a deliverable — a concept, a recording, ideally something project-based.
• Ukrainian Web3 talent network (~1000 people / entities) is the seed audience. Students can get routed into companies via the same flow.

Reiner's framing: this is effectively a Web3-based parliament — any engaged citizen contributes; AI stitches circle outputs into a forward-looking read on "what is Ukraine becoming." Could be sold into policy / lobby channels that need that forward view. Also: candidate-seeking entities belong in the pool too — it's a talent market, not just a supply side.

Demo 2 — AI call assistant with video overlay

Max ran a working build on the live call. It took a while to compile and hit some camera-source bugs, but the core worked:

• Virtual camera selected → outbound video carries an extra overlay layer.
• Overlay elements: a blue action button + an element tracking Max's face (followed semi-correctly).
• Output layer can carry anything dynamic — timer, speaking-time, detected intention, pulse, hand-gesture prompts, text, speech bubbles for non-speaking participants.

Key architectural point (Reiner re-confirmed twice in different words): the tool captures the camera + audio stream at the device level, so it's Teams/Zoom/phone-agnostic. No plugin per platform. Installed per-device. Cloud API can host reference data (e.g., HGL) rather than shipping it locally.

Framing that landed

• "Real-time feedback through images" — not a sidebar chat, visual output inside the call.
• "A live chat, but visualised / video." People stay immersed in the video instead of reading.
• Worked example (Reiner): "Reiner is drifting off Agenda 1 into Agenda 0.5" appearing as an overlay — though he flagged it could be intrusive.
• HGL can be embedded on demand, triggered by a button.
• Games / constrained dialogue formats work on the same primitive (e.g., a speak-without-"I" game where using "I" triggers a visible crash).

Which product, for whom — use-case discussion

Reiner pushed hard: "We are close to a gold mine but the nuggets aren't in our hands." Time to pick one niche and make it tangible.

Candidates surveyed:

• Weekly management / team meetings — dreadful for most; if they become 10 min shorter and more alive, clear value.
• Army daily briefing — structured, ritualised, mobilising. Same infrastructure, different specification.
• Customer service / sales conversations — already recorded, replicable, metrics are obvious (conversion, NPS). Reiner's vote: start here.
• Coaching-style conversations — tool surfaces what's between the lines.
• Lawyers preparing a pledoie, and other structured ritualised speaking.

Reiner's structural view: infrastructure is generic; each specification produces a context-specific service. Ship one specification well → expand.

The Dojo frame

Reiner has already been using the word "dojo" with practice clients without the AI. A recent session with a CEO client crystallised the design:

The client said directly that she could not, in real time, be non-authoritative — it's hardwired. She's willing to practise in a dojo, but knows her default style won't flip mid-conversation. Her own resolution: delegate the facilitative style to another leader in the team.

Resulting design principle: each participant can have their own companion, showing what matters to them, invisible to others. Analogy that stuck — two chess players, each with their own assistant program, still playing the real move themselves. This also sidesteps GDPR pressure: if I'm the only one seeing my feedback, the observation problem shrinks.

Connecting this to the latest work on group learning: AI as observer, group rotating roles, no right/wrong — goal is to expand each participant's behaviour repertoire and expose them to how others think. Dojo = a space to expand repertoire.

Fintech sales case

A current client of Max's: a fintech selling POS terminals to cafés and shops. ~30 salespeople now, plausibly 300+ soon. System auto-dials from yellow pages, routes live answers to whichever salesperson picks up; seniors intervene when escalation is needed. No call recordings — the sales director can't teach or monitor effectively. Reiner's read: highly Zurich-shaped, immediate metrics (conversion rate, cold-call success).

Conclusion on niche

Customer service / sales is the first target — highest replicability, already-recorded territory, clear metric, every org has them, none of them treat it as luxury. Zurich stays as the proof-point in the background.

Closing agreement — demo video

Two external partners (a Phoenix Consultancy contact and the business connector above) both explicitly want a video/demo to forward. Decision: ship a short scenario-based demo film, not a product.

• 2-minute sales conversation that ends badly.
• Same conversation again, with AI giving signs / overlays during the flow.
• Second run ends well. Shows the real-time-feedback distinction we own.
• Can be built with AI generation rather than shooting live.
• Purpose is to open the door — "give us your pain points, we'll build the product."

Operations — away from Notion

Max has stopped using Notion; flagged cancelling the licence. Uses Antigravity (dev environment with Claude Code under the hood) as the working surface. Chat-driven development, folder-based memory of project docs — everything we had as Notion pages can become actual features: smart notifications, per-participant email summaries after each call, LinkedIn drafts, demos embedded on the site. Notion = "too boring."

Plan: new site assembles the scattered pieces — meeting notes, demos, experiments — plus an embedded smart assistant. Reiner: once we have a demo or video, be bold enough to post to LinkedIn; people will ask and we need something to hand them.

Action items