Digital human
Heliosync’s long-term digital-human and artificial-cognition initiative. A character with a real audience today; a persistent digital entity, and a much harder question, tomorrow.
The objective is not another chatbot and not another AI avatar. It is an entity that continues — one that remembers, holds an identity, understands where it is and acts inside limits it did not write for itself. None of that exists yet, and this page is explicit about which half is which.
The problem
Current digital characters and AI influencers are largely dependent on humans, or on predefined systems, to create posts, respond to people and perform activities. The models underneath them are capable; what they lack is anything that continues from one request to the next.
What we are building
Shanaya is Heliosync’s long-term digital-human and artificial-cognition initiative. The objective is to build a persistent digital entity rather than another chatbot or AI avatar — which means separating the foundation model from the systems that give an entity continuity: memory, identity, a world model, agency, tools and learning.
The longer aim is to move digital personalities from static content creators into persistent, intelligent digital entities — and, further out, to ask whether that path leads from artificial intelligence toward artificial cognition. That is a research question, and this page treats it as one.
Where she is today
Shanaya publishes on Instagram as @itsokshanayaa (opens in a new tab). Every reel is currently scripted, produced and posted by a person. She is a character we operate, not yet a system that operates itself — and we would rather say so than imply otherwise.
The account is operated by StudioNoir, an independent creative brand. Heliosync’s work is the digital human system described below — which does not exist yet.
The architecture
A foundation model is the reasoning substrate, and it is the part of this that already exists. What it cannot do is continue. It keeps no memory of yesterday, holds no stake in being the same thing tomorrow, and carries no picture of the room it is standing in. Everything that makes something an entity rather than a service sits around the model, not inside it — which is why the programme is seven systems and the model is only the first of them.
Status
None of this exists yet. The architecture below, the objectives after it and the research question at the end of this page are statements of intent, not of capability. This is an active research programme with no release date attached. Shanaya cannot currently reply to messages, remember anything, publish on her own or take instructions — and we will say so here until she can.
The substrate
The reasoning layer — language, inference, and the ability to work a problem through. It is the part that already exists, and on its own it is not enough: a model that reasons brilliantly still begins every conversation as a stranger.
What persists
A durable record of what actually happened, held across sessions and across months, so that experience accumulates rather than ending with the context window it was formed in.
Who it is
A self that survives the conversation — consistent values, voice, history and preferences — so the entity answering on Tuesday is recognisably the one that answered on Monday.
Where it is
An internal representation of its own situation: what is true, what has changed, who it is speaking to and what follows from that — maintained, rather than reassembled from whatever the prompt happened to mention.
What it wants
The capacity to hold an objective and pursue it over time, choosing the next step rather than waiting to be told it — and stopping at a boundary the system enforces rather than one the model is asked to respect.
What it can touch
Access to real software through sanctioned interfaces, because an entity that can reason about the world but cannot act in it is a conversation rather than a participant.
How it changes
Change that persists — behaviour updated by outcomes and correction, so the system is different after an experience than it was before it, and stays different.
The objectives
None of these is a capability Shanaya has. They are the specification the programme is measured against — the ten things a persistent digital entity would need to do before the word entity is earned. The distance between this list and what runs today is the work.
To work a problem through rather than pattern-match to the shape of an answer — and to be legible about how it got there.
To recall what happened, not merely what was said in the last few thousand tokens, and to forget on request.
To remain one entity across conversations, platforms and months without being re-described at the start of each.
To register where it is, what is happening around it, and what has moved since it last looked.
To hold a picture of how things stand and revise it when the facts change, instead of treating each prompt as the entire world.
To operate real software through sanctioned interfaces — the same tools a person would use, under the same permissions.
To act only inside an explicit allowlist, where the limit is enforced by the surrounding system and not by the model's good intentions.
To carry an objective across sessions and make progress against it without being re-prompted at every step.
To improve from outcomes rather than resetting to the same defaults after every correction.
To be one entity across devices, platforms and time — not a fresh instance wearing the same name.
The hard problems
A list of components is not a system. These are the four places where we do not yet know how to build the components well — a demo is a graphics problem, and a digital human that runs unattended is an engineering and safety problem.
Memory and identity are easy to list as components and hard to keep. A character who contradicts herself across a hundred conversations stops being a character, and memory that survives months is a different engineering problem from context that survives a session.
Presence is a timing problem before it is a graphics problem. The pause a person reads as thought is short, and everything — speech, language, expression — has to complete inside it.
An agent that reads public messages and can also act on them can be told what to do by anyone who messages it. Separating untrusted input from anything with write access is a prerequisite, not a hardening step.
Autonomy that gets the account banned is not autonomy. Anything automated has to run through sanctioned APIs, within their messaging rules, with AI disclosure handled properly.
The path
Three stages, and the first one is where we actually are. Nothing moves to the next until the safety work for it has been done and shown to hold.
Today
Shanaya is written, produced and published by a person. The character is live; the system is not.
Next
Drafting and scheduling under human approval — the model proposes, a person still presses publish. This is where the safety work is proven before anything is trusted to act alone.
Later
Replying, deciding and creating unattended, inside the bounded actions described above — on her own behalf, or on behalf of a person or brand she works for. Long-horizon, and dependent on the problems above being genuinely solved.
StudioNoir
The nearest practical use of this work is the one already visible: a digital personality a brand can work with — an AI influencer. That is real work with a real audience, and it runs through StudioNoir, whose brief is realism. Lighting, motion and performance are held to a standard where the AI is not the point and the character is.
Today the role is carried by people. Every post is written, produced and published by the team at StudioNoir; the character is theirs, and so is the craft. What is being built at Heliosync would let a digital human carry that role itself, and an influencer is one of the simpler things a persistent entity could be pointed at.
Which sets the relationship the right way round. StudioNoir is one practical application of the digital-human technology. Shanaya is the research project underneath it — the application is how the work meets the world early, and the research is why the work exists at all.
The research question
Everything above is engineering, and engineering is not why it is being attempted. A model today answers and then stops existing. Between one request and the next there is nothing there: no memory of what happened, no position in the world, nothing that carries over. Memory, identity, a world model, agency, tools and learning are the systems that would close that gap — and closing it raises a question that the engineering itself does not answer.
Can engineered systems move from artificial intelligence toward artificial cognition — and if they can, could such a system come to exhibit properties that are relevant to consciousness?
That is a question we are asking. It is not a result we are reporting, and the distinction matters more here than anywhere else on this site. Heliosync has not built consciousness. Heliosync does not claim to be close to building it, and does not claim to know whether it can be built at all. Consciousness remains an open problem in neuroscience and in philosophy: there is no agreed definition of it and no accepted test for it, which means no engineering team — ours included — is currently in a position to demonstrate it in a machine.
What can be investigated is narrower and more tractable. If a system remembers, models its own situation, holds an identity across time and acts toward goals, it becomes possible to study carefully whether it begins to show any of the properties the literature treats as relevant to cognition — and, just as usefully, where it plainly does not. Negative results count. A finding that these properties do not emerge would be worth as much as a finding that they do, and it is the more likely of the two.
That is the standard we hold this to: an open question, pursued in public, with the claims kept behind the evidence rather than ahead of it.
To be unambiguous
Heliosync makes no claim to have created consciousness, no claim to be near to creating it, and no claim that Shanaya is conscious, sentient or self-aware. She is not. Shanaya today is a human-operated character on Instagram; the architecture described on this page has not been built; and the question above is a research direction rather than a finding. We will describe it as nothing more than that until there is evidence that says otherwise.
Collaborate
Shanaya works with brands today through StudioNoir — that is a creative conversation and it goes to them.
If you work on memory and continuity, agent safety, real-time multimodal systems or the harder question of machine cognition, the conversation about the underlying research goes to us.
Work with us
Partnerships, research collaboration, investment, or a role on the team — the door is the same one.