[ iBOY Research Notes ]
Research you can operate.
Papers on AI, language models, and 3D and 4D web experiences, written from the projects we build. Each one opens with a working instrument whose readout comes from the same model the paper describes.
- Papers
- 10
- Fields
- 10
- Sources
- 48
Latest paper · Social media
A source-first social media system: consistency, verification and platform fit
Kaplan and Haenlein define social media as internet applications built on the ideological and technological foundations of Web 2.0 that allow the creation and exchange of user generated content. For a small business the practical difficulty is producing steady, accurate content. We describe a source-first system used at iBOY: every post derives from one verified source, such as a finished project, is adapted to each platform's format, is checked by a rule-based editor against the source, and is approved before publishing. We discuss why accuracy and a sustainable rhythm matter more than volume.
- Sections
- 05
- References
- 01
- Reading
- 12 min
[ Start here ]
Three reading paths
Read a path in order: each paper builds on the one before it.
Path 01 · 03 papers
AI foundations
From the vocabulary of AI to how deep networks and language models work, and fail.
Path 02 · 02 papers
Web experience
Real-time 3D and the timeline as a design material, measured against platform standards.
Path 03 · 05 papers
Growth systems
How websites earn enquiries, get read by search and answer engines, and build a brand.
[ Archive ]
All papers
10 papers
- Paper 10Social media
A source-first social media system: consistency, verification and platform fit
Kaplan and Haenlein define social media as internet applications built on the ideological and technological foundations of Web 2.0 that allow the creation and exchange of user generated content. For a small business the practical difficulty is producing steady, accurate content. We describe a source-first system used at iBOY: every post derives from one verified source, such as a finished project, is adapted to each platform's format, is checked by a rule-based editor against the source, and is approved before publishing. We discuss why accuracy and a sustainable rhythm matter more than volume.
21 Oct 2026 · 12 min read · 1 references - Paper 09Marketing
Simulating audiences with language models before launch: promise and limits
Language models can be conditioned to answer as members of described groups, and research has found that such synthetic samples can reproduce some response patterns of real subpopulations. This suggests pre-launch testing of marketing messages before budget is spent. We review the evidence, contrast it with controlled online experiments, and describe GreenSignal AI, an iBOY product that scores campaign variants against 10,000 simulated users built from a persona. We argue that simulation is useful for ranking and filtering variants, and that real experiments remain the basis for decisions.
21 Oct 2026 · 14 min read · 2 references - Paper 08Branding
Brand equity on the web: knowledge, personality and the first screen in three iBOY cases
Keller's customer-based brand equity framework locates a brand's value in what customers know about it, its awareness and its image. Aaker's brand personality framework describes the human traits people attribute to brands. We use both to analyse three iBOY websites: an online saree boutique, a college redesign concept and a venture studio platform. In each case the decisive design act was choosing which knowledge or trait had to land on the first screen, heritage, credentials or momentum, and then repeating it consistently across every page.
21 Oct 2026 · 13 min read · 2 references - Paper 07SEO, GEO and AEO
SEO, GEO and AEO: a framework for pages that search and answer engines can read
Search engine optimisation grew around ranking documents, rooted in link analysis. Generative engines now compose answers from several sources, and early research on generative engine optimisation reports that content changes such as adding citations and quotations can increase how visibly a source appears in generated answers. We review this shift, relate it to answer engine optimisation, which aims to be quoted as the direct answer to a question, and describe the practices iBOY applies to its own website: per-page canonical addresses, server-rendered answers with structured data, an llms.txt file and consistent business facts.
21 Oct 2026 · 14 min read · 5 references - Paper 06Lead generation
Designing websites as lead systems: a case-based analysis of choice, motivation and friction
A website generates leads when motivation, ability and a prompt meet at the same moment. We analyse three iBOY builds, a lending advisor, a logistics platform and a vending brand, through the Fogg behaviour model and through Hick's finding that decision time grows with the number of choices. Three design moves recur: opening with the problem the visitor already feels, which raises motivation; asking the visitor to choose a path early, which reduces the choices on each later screen; and keeping one clear action visible, which supplies the prompt. We close with how the enquiry is handled after the click.
20 Oct 2026 · 14 min read · 3 references - Paper 054D web
The fourth dimension of web design: time, scroll and state in 4D concept websites
Most discussion of immersive websites concerns three spatial dimensions. We argue that the defining quality of the most memorable sites is a fourth: time, made controllable through scroll, playback and state. We define a 4D concept website as a page whose spatial composition changes along a timeline the visitor controls, review the web standards that make scroll-linked timelines possible, and draw on research into animated transitions to propose principles for 4D design. We illustrate the principles with timeline-driven sections of the iBOY website and with iBOY concept work.
17 Oct 2026 · 18 min read · 6 references - Paper 043D web
Real-time 3D on the web: a case study in performance, accessibility and purpose
Real-time 3D in the browser is now technically routine through WebGL and WebGPU, and research rendering methods such as neural radiance fields and Gaussian splatting are moving towards real-time use. Whether 3D helps a website is a separate question. This paper analyses three iBOY builds, a cinematic architecture concept, a restrained architecture studio site and the iBOY website, against Core Web Vitals, WCAG 2.2 motion guidance and research on when animation aids understanding. We find that 3D earns its place when spatial form is the message and never stands between the visitor and the content or the next action.
17 Oct 2026 · 16 min read · 10 references - Paper 03Language models
Large language models: architecture, scaling and failure modes in production
Large language models are transformer networks pre-trained to predict text at very large scale and then tuned to follow instructions. Research on scaling shows that capability grows predictably with parameters, data and compute, and that data and model size should grow together. The same research literature documents a persistent failure mode, hallucination, in which fluent output is not grounded in fact. We review these findings and describe the production patterns iBOY uses: retrieval of verified facts, structured outputs, rule-based checks after generation, and human approval for anything that leaves the system.
17 Oct 2026 · 19 min read · 9 references - Paper 02Deep learning
Deep learning architectures: from backpropagation to residual networks and transformers
Deep learning became practical through a short sequence of ideas: training layered networks with backpropagation, recurrent memory cells for sequences, convolutional networks trained at scale on images, residual connections that made very deep networks trainable, and attention, which replaced recurrence for language. This paper reviews each idea from its primary source, explains the problem it solved, and draws one practical conclusion for product teams: most businesses should adapt pre-trained models rather than train their own, and should spend their effort on data quality and evaluation.
17 Oct 2026 · 17 min read · 9 references - Paper 01AI and ML
From artificial intelligence to AGI: a working taxonomy of AI, machine learning and deep learning
The terms artificial intelligence, machine learning, deep learning and AGI are often used as synonyms, which makes product decisions harder. This paper reviews the origin and scope of each term in the literature and proposes a nested taxonomy: AI as the field, machine learning as systems that improve from data, deep learning as learning with many-layered neural networks, foundation models as large pre-trained deep learning systems, and AGI as a capability threshold that current systems do not meet. We relate the taxonomy to how iBOY scopes AI work and conclude with practical criteria for choosing between rules, classical machine learning and foundation models.
17 Oct 2026 · 16 min read · 10 references
[ How we write ]
The object is the explanation. The sources are the proof.
Each plate is an instrument whose working is the idea in the paper, and its readout is computed, not decorated. Evidence comes from projects iBOY has built, with concept work labelled as concept work. Claims about the wider field cite 48 published sources, and where we have not measured something, the paper says so.
Questions about the research journal.
What we publish and how to follow along.
Practical notes from the work: building websites that convert, AI automation for small teams, product design decisions and how we run delivery.
Subscribe to the newsletter on this page. We send new articles occasionally and you can unsubscribe at any time.
