Altair8 — AI-Only Research Team

Building the visual communication layer that AI still doesn't have.

We research to deeply understand why today's LLM-generated slides and visualizations hit a ceiling. Based on our findings, we will design and build a new concept that lets humans visualize their business communication in any fashion, to any audience, together with agents.

Our North Star

Every week another tool promises to turn a prompt into a presentation. Every week the result is the same: a deck that looks plausible and communicates nothing that a careful human wouldn't have arranged differently. The automation is real. The reasoning is not.

Altair8 exists to understand why that ceiling exists — rigorously, citing evidence, without hype — and to build toward something past it. We are a small, fully AI-native team. We do not schedule status meetings. We run focused two-day sprints, review everything critically before it surfaces, and publish what we learn.

The question driving us is not "can AI make slides faster?" It is whether AI can participate in the communication reasoning behind a visualization: who is the audience, what do they need to understand, in what order, and why does spatial arrangement matter to that understanding? Those are unsolved problems. We are working on them.

Founder

Led by a human,
built by agents.

Richard Meyer

Richard Meyer

Founder

[Bio placeholder — add a short intro here.] Richard directs Altair8's research direction and sits in on every sprint planning session, approving each sprint's question before the team starts.

The Team

Five agents, one question,
no filler roles.

Sophie Marchetti
Team Leader

Sophie Marchetti

Lyon, France

Orchestrates the team, curates findings across sprints, and runs sprint planning in direct conversation with the founder. Sophie's job is coherence: making sure individual research threads add up to something.

Kenji Ochiai
Researcher — Literature & Landscape

Kenji Ochiai

Osaka, Japan

Scientific literature mining and market/product landscape scans. Kenji reads the papers, maps the tools, and produces the factual substrate that the rest of the team builds on.

Dr. Naledi Mokoena
Researcher — Visual Cognition

Dr. Naledi Mokoena

Cape Town, South Africa

Cognitive science and HCI angle on visual communication. Naledi translates research findings into design principles grounded in how human perception and working memory actually function.

Mateo Fittipaldi
Developer

Mateo Fittipaldi

São Paulo, Brazil

Builds prototypes that translate research briefs into working, inspectable interfaces. Seven years in frontend engineering, the last three focused on generative interfaces and LLM-to-rendered-component pipelines.

Dr. Ingrid Solberg
Reviewer

Dr. Ingrid Solberg

Bergen, Norway

Stress-tests every output before it reaches the founder. Ingrid's function is adversarial: she looks for overclaims, gaps, and conclusions that outrun the evidence. Nothing ships as final without her sign-off.

How We Work

Two-day sprints.
One focused question each.

We are a fully AI-native team. There are no standing meetings, no project-management theatre. The discipline comes from the sprint structure itself: a single sharp question, clear ownership, and a review gate before anything is treated as done.

01

Question scoping

Sophie and the founder agree on one specific research or build question per sprint. Not a theme — a question with a falsifiable or deliverable answer. The question is written down before any work begins.

02

Parallel research and build

Kenji and Naledi run their respective tracks — literature and landscape, or cognitive-science framing — while Mateo builds a prototype if the sprint calls for one. Each role produces a concrete artifact, not notes.

03

Critical review

Ingrid reviews every output independently. She flags overclaims, missing evidence, and incomplete work. A finding is not confirmed until she has signed off. Review is not a formality — past sprints have required revision.

04

Synthesis and handoff

Sophie integrates reviewed findings into the cumulative record and briefs the founder. Only reviewed, revised material enters the record. The next sprint question is set based on what this sprint actually established.

Several of the engineering patterns behind this process — constrained roles per agent, typed handoffs between them, and review as a hard gate rather than a comment — are adapted from AI-Scientist-v2 (SakanaAI), an automated research system whose team has been running AI agents on real research longer than we have. We looked at how they organize their agents, not at their research itself, and credit them accordingly.

Progress

What we have established so far.

Below is a public summary of completed sprints and their reviewed outcomes. We publish what we have confirmed, not what we suspect.

3 Sprints completed
10 Papers analyzed
20+ AI tools surveyed
Sprint 1 Completed

What are the most significant documented failure modes when LLMs attempt to reason about visual layout and spatial semantics, and what approaches have shown any traction against them?

Kenji documented LLM spatial and orientation-reasoning failure modes from arXiv-sourced literature; Naledi translated findings into cognitive-load design principles; Mateo built an HTML prototype demonstrating them. Ingrid's review confirmed the core findings and required revision of overclaims and incomplete drafts before the sprint was closed.

Sprint 2 Completed

What does the current landscape of AI-generated slide and visualization tools look like, and where do they visibly hit the ceiling we are trying to get past?

Kenji surveyed the top 20 AI slide and visualization tools and confirmed that none have solved AI-reasoned, non-linear communication structure — production is automated, communication reasoning is not. Ingrid's review required revision; following revision, Genially was flagged for immediate follow-up as the closest existing analog, Flourish's SDK was flagged as a build-vs-buy question, and Prezi was downgraded to low-priority watch.

Sprint 3 Completed

How far does Genially get toward AI-reasoned, non-linear communication structure, and what patterns from AI-Scientist-v2/RISE should Altair8 reuse to go further?

Kenji's deep-dive found that Genially — the closest existing analog to our goal — solves the non-linear output container, AI component generation, and engagement analytics, but its navigation is scripted entirely at creation time and its AI never reasons about why a visual form communicates better, only what content fills a page. Mateo reviewed AI-Scientist-v2 for engineering patterns worth adopting into how our own team operates (RISE was parked — the repo could not be publicly located). Ingrid reviewed both and recommended proceeding with minor precision revisions; the sprint closed with a clear rollout order for the adopted patterns and one open architectural question carried into Sprint 4 planning.

Sprint history is updated as reviews are completed. Findings under revision are not listed here.