Author: Dariusz Doliński (Darkar Sinoe), Founder & Semantic Architect | Synthetic Souls Studio

Author: Darkar Sinoe | Synthetic Souls Studio™

Document Type: Strategic Implementation Protocol (White Paper)

Date: March 2026

Status: Classified / Strategic Asset

The Polish e-commerce market faces an architectural decision whose consequences will unfold over the next five years. This is not a decision for government, regulators or investment funds. It is the decision of every merchant who this week receives an invitation to the Von Halsky partner program.

* 15.1–15.2 million are active buyers (at least one transaction in period). Total registered Allegro accounts exceed 20 million. FY2025 operating data.

Before comparing the architectures of Allegro and Von Halsky, we must establish what we mean by purchase intent. This is not a philosophical question. It is an engineering question — because the answer determines which level a given system operates on, what it sees, and what it cannot see.

Definition: the consumer knows what they want and can name it precisely.

This is the layer every search engine has operated on since 2000. Google built an empire on it. Allegro built PLN 70 billion GMV on it. Von Halsky built its revolution narrative on it. Architecture: query → token parsing → database matching → ranking → result. None of these steps requires understanding the human. They require processing text.

Definition: the consumer knows what they need but cannot name it precisely.

Allegro does not handle this at all. Von Halsky tries — but Bielik v3 operates exclusively on the closed catalog of 4,000 InPost partners. It will not find the best gift for mom. It will find the best garden product available from InPost partners. That is not the same thing. The difference is structural, not qualitative.

Definition: understanding who the human asking the question actually is. Not what they typed into the chat window. But: what is their archetype, what semantic language is natural to them, what drives their decisions, why are they searching NOW.

No deployed commercial system in Poland operates at Level 3. Not Von Halsky. Not Allegro. A language model alone does not operate at Level 3. It requires a doctrinal layer — a methodology sitting above the model, imposing archetype frameworks upon it.

In July 2025, Soumyadeb Mitra, CEO of RudderStack, reached out to the author of this audit first. In his email he wrote: “Your feedback as an AI leader would be invaluable for us.” RudderStack builds Customer 360° — a full picture of the customer assembled from transactional and behavioral data. Bought this, clicked there, abandoned cart here. A sum of the past, not an understanding of the future.

Four months later Mitra was already writing about “intent observability” — asking: “Did the agent understand this intent and act accordingly?” This is a pivot: from data about behavior toward understanding of intent.

On 26 March 2026 Mitra published a post about the death of SaaS and the commoditization of interfaces. Under the same post Darkar Sinoe commented, applying the thesis directly to Von Halsky and Allegro. Mitra publicly liked that comment — a gesture confirming that the analytical direction is aligned with what the data infrastructure leader sees as the future. This is not coincidence. It is the same trajectory seen from two sides of the architecture.

“This is an absolutely revolutionary form of shopping that no one has yet created in such a comprehensive way...” — Rafał Brzoska, CEO InPost, Von Halsky presentation 2026. Sentence unfinished. PR without conclusion. Revolution without defining what changed technologically.

Von Halsky is powered by Bielik — a Polish language model by the SpeakLeash foundation (ACK Cyfronet AGH), built for Polish technological sovereignty. A valuable project for the country. But Von Halsky is a commercial product serving 16 million users in purchases requiring intent understanding. And here a problem emerges that is not an opinion — it is a benchmark.

Architectural conclusion: the model that last week ranked last in Polish text proofreading and 8th in citing national literature powers a system meant to “understand” ambiguous, emotional, contextual purchase queries from a worried parent, an elderly person seeking medication, a woman buying a gift for her grandmother. It will not manage. It will return the most popular result from the database, wrapped in a polite prompt. This is an upgraded search engine.

Test authors note that Bielik “fits well into RAG (Retrieval-Augmented Generation) architecture.” RAG retrieves from an external database and generates a response based on it — exactly what Von Halsky does on the catalog of 4,000 partners. Brzoska sold RAG on products as a revolution. Prof. Kawa does not call it a revolution.

Allegro is a reactive system. It captures demand once the customer already has formed purchase intent and is searching for a specific product. This is a precise definition — and simultaneously Allegro’s strength and its fundamental limitation.

Allegro does not see pre-intent. It does not know someone is thinking about a purchase before they open the platform. This is exactly where InPost tries to strike — Von Halsky should intercept the customer before they open Allegro. Strategically sound observation. The error is that Bielik as RAG on a closed catalog captures intent at the same level as Allegro — just in a different interface.

Allegro holds purchase data on 15 million Poles over more than a decade: transaction history, abandoned carts, ad clicks, basket values, reviews, returns. This is the foundation for Customer 360°. InPost holds logistics data — who sent, who collected, from which locker, in which neighborhood. Data about the final phase of purchase. Not about intent.

Prof. Kawa, 26 March 2026: “Will the user actually see the best product for them, or the one best positioned within the ecosystem? How objective will the recommendations be? Will the purchase path be designed to steer the customer toward a faster decision — using dark patterns? UOKiK is already developing AI tools for detecting dark patterns.” These are not hypothetical questions. UOKiK enforces its answers with fines.

Global ICPEN/GPEN sweep, January–February 2024: UOKiK examined 10 Polish platforms — 50% employed at least one dark pattern. Globally (642 platforms): 75.7% employed at least one. UOKiK is testing GPT-4 for automated dark pattern detection. Production deployment in preparation.

The key question: when two Von Halsky partners offer identical products at identical prices with identical delivery — what determines the order of recommendations? InPost has not published ranking criteria. There is no public algorithm document. There is no appeal mechanism for merchants. InPost earns on parcels passing through lockers — not on recommendation accuracy. This is a structural conflict of interest built into the business model.

You receive an email from InPost: Von Halsky, zero commission, 16 million users, free delivery for customers. This is an offer you cannot refuse — or you can, if you understand what you are signing.

Access to 16 million InPost Mobile users. Product visibility in Von Halsky. Free delivery and returns for customers within InPost+. No “classic commission on sales” for now. A new customer acquisition channel without building your own AI. These are real benefits — honest at the launch stage.

Your customers’ purchase data stays with InPost. When a customer buys your product through Von Halsky — InPost knows what they bought, from which locker they collected it, what other products they browsed before choosing yours, and which neighborhood they live in.

When Von Halsky reaches 10,000 partners — InPost negotiates the terms, not the merchant. “For now” are two words placed deliberately.

In the Von Halsky model the customer buys in the InPost app, pays via InPost Pay, collects from InPost locker. The merchant provides the product — everything else belongs to InPost. The customer does not know they bought “from your store.” They know they bought through Von Halsky. Next time they will return to Von Halsky — not to you.

You escape the Allegro cage (commissions and advertising) and enter the Von Halsky cage (customer data and customer relationship). The cage is different. It is equally locked.

Von Halsky processes a query. The customer types “sports shoes for running” and the system retrieves products from the category and ranks them. This is processing query. The system processes text. The system does not process the human.

Why is someone searching for running shoes right now? Option A: just started running, is a beginner, fears injury. Option B: runs for 10 years, looking for a specific new model. Option C: buying a gift for a partner after a medical diagnosis. Option D: a coach searching for shoes for an athlete with pronation. All four people may type identical queries. Processing query gives all four the same result — the most popular model in the category. Understanding human gives each what they actually need.

Model power alone is insufficient — but it matters as a foundation. GPT-4o/GPT-5 (OpenAI): multi-turn context understanding, emotional state inference — potential Level 2–3 with a doctrinal layer. Claude (Anthropic): high semantic precision, good tone stability — potential 2–3. Gemini 2.x (Google): open internet access and multimodal data — potential 2–3. Bielik v3: RAG on closed catalog, Level 1 and partial 2 — confirmed by March 2026 benchmark, 12 models, bottom of ranking.

Syntax Protocol™ is a deterministic semantic architecture developed by Synthetic Souls Studio™ in 2025, documented through 47 test sessions (Protocol 07, July–August 2025). The principle: 1 intention / 1 generation / 0 post-production.

In practice, this means the system does not select the product with the highest purchase probability — as every RAG and every recommendation algorithm does, whose sole optimization criterion is conversion maximization, not archetypal adequacy. The system constructs a purchase path appropriate to the customer’s archetype: without averaging, without probabilistic guessing, without generating a “safe” middle-ground answer. It classifies archetype → maps to product semantics → generates a recommendation that answers “why are you searching”, not only “what did you type.” That is the fundamental architectural difference between a doctrine and a tool.

Attempts to copy, acquire, or reverse-engineer Intent Architecture and Human360°™ have already been made by global players with the highest capital. Including entities operating at the top tier of the luxury sector — a category where the stakes are a billion-dollar market and heritage spanning more than a century.

All of these attempts ended in failure and technological dead ends. Not due to lack of computational resources. Not due to lack of access to language models. They had the infrastructure. They had the budgets. They lacked the doctrine.

The corporation that sends engineers to “deconstruct” this architecture will receive on output: an expensive proprietary RAG built on their own database, wrapped in prompt engineering and called “intent architecture.” Exactly what they already have — and which already underperforms. They will burn the budget. They will lose the time. And in 18 months they will return to the same point: data without doctrine.

The solution exists. The road to it does not lead through the R&D department. It leads through licensing the methodology or direct collaboration with its creator. Every other path is purchasing an expensive shell.

Allegro has the data. It does not have the doctrine. InPost has the distribution. It does not have intent architecture. Bielik has the ambition — and ranked at the bottom of 12 models on its own linguistic home turf in March 2026. Mitra sees the direction. Kawa says: “I wouldn’t call this a revolution.” Both are right at their respective levels.

The real revolution in e-commerce is not a chatbot. It is a system that understands why a human buys — and delivers it before they manage to formulate the query. Nobody in the Polish market has this. But someone has already described how to build it.

Glossary of Terms (Dictionary of the Third Era)

Biological AI Cinema™ — a film production methodology based on simulating biological truth in latent space. Result: completion rate 21–36% vs. industry average 4–8%.

Syntax Protocol™ — a deterministic operating system for visual production. Shooting ratio 1.5:1. Zero post-production. Identical result across 6 AI models.

Biological Governor — Layer L2 controlling the physics and biology of generation: SSS, muscle tension, saccades, fabric physics.

Temporal Coherence Optimization — technology maintaining visual stability for 30–120+ seconds (vs. standard 5–10 sec.).

Soul Gap — a measurable disproportion between the technical correctness of an image and its inability to trigger biological resonance.

Smoothing Bias — a systemic error in diffusion models consisting of the elimination of biological micro-details (pores, asymmetry, tremor) which the viewer's brain interprets as evidence of life.

SDR (Semantic Density Ratio) — a content semantic density indicator. Market standard: < 0.2. Syntax Protocol™: > 1.5.

Embodied Simulation™ — a technique in which AI does not "draw" emotions but simulates an emotional experience internally, resulting in the emergence of micromimicry and asymmetry.

Neural Cinematography — engineering of camera parameters (angle, depth of field, motion) directly within latent space, not as a post-production effect.

Aether Skin Protocol™ — a rendering sub-layer for the Beauty sector, introducing controlled biological micro-imperfections (pores, perspiration, blood vessels) that eliminate the Uncanny Valley.

Darkar Sinoe (Dariusz Doliński)Semantic Architect & AI FilmmakerFounder, Synthetic Souls Studio™ | Talent Guide @ BlueFoxes ParisCreator of The Syntax Protocol™ | Era III Doctrine

→ Dictionary of the Third Era: syntheticsouls.studio/dictionary-of-the-third-era→ Film Gallery: syntheticsouls.studio/gallery-2→ Contact: syntheticsouls.studio/contact-2

LEGAL NOTICE

Syntax Protocol™, Biological AI Cinema™, Semantic Fortress™, Semantic Steering Layer™, Aether Skin Protocol™, Human360°™, Emotion Architecture™, Embodied Simulation™, Neural Cinematography™, Era III™ and Soul Gap are registered designations of Synthetic Souls Studio™ (Dariusz Doliński). All rights reserved.

The methodology, production architecture, prompt structures and internal audit tools described in this document constitute the intellectual property of the author and are protected by copyright. Reproduction, citation or commercial implementation without written consent is prohibited.

© 2025–2026 Synthetic Souls Studio™. Dariusz Doliński / Darkar Sinoe. All rights reserved.

Reference video material:

Human360° | From Data to Humanity | AI Storytelling by Darkar Sinoe | Synthetic Souls Studio

Watch on YouTube

Copyright © 2025 Darkar Sinoe & Synthetic Souls Studio™. All rights reserved.

The methodologies Human360°, Imprint™, Semantic Steering Layer™, and Soul Gap™ are the intellectual property of the author.

Ready For A Revolution In Film

Advertising?

→ Schedule a Free Consultation (20 min) write → Watch the EVELLE Film → Go to the contact form write

About the Author

Dariusz Doliński (Darkar Sinoe)Semantic Architect | Founder, Synthetic Souls Studio™

Creator of Emotion Architecture™ and Human360°, AI storytelling methodologies achieving 28–36% completion compared to <10% market standard. 13 years of experience in digital creation, 11 months of research in AI-driven narrative intelligence.

Officially recognized by Google Knowledge Graph as the originator of the concept of intention as a semantic driver in AI filmmaking.

Flagship Projects:WELES (11-min AI cinema) • AETHER (luxury beauty transformation) • EVELLE (case study)

Headquarters: Warsaw

Collaboration: Dubai • Mumbai • Los Angeles📩

darkar.sinoe@syntheticsouls.studio📞 +48 531 581 315

info@syntheticsouls.studio

++48 531 581 315

© 2025 Copyright By Synthetic Souls Studio All Rights Reserved