Research
We explore novel mathematical paradigms for machines to learn, remember and reason with pure 3-dimensional architectures.
Build
We train and architect our own foundation models from scratch, ensuring total algorithmic sovereignty.
Integrate
We embody intelligence into real-world operating systems: acoustic voice, continuous vision, and physical automation loops.
“We are not building a product.
We are building a mind.”
Three Directions.
One Sovereign Goal.
Exploring conceptual machine intelligence from first principles. We work in quiet depth, focusing on pure 3D continuous memory, sovereign compact models, and ambient sensory embodiment.
ARIA — Pure 3D Frequency Architecture
Conventional AI relies on complex transformer blocks and artificial vector spaces. Our AI does not work on that at all — we have ONLY 3D. We are building machine intelligence in a totally new way, architected exclusively within a 3-Dimensional frequency continuum.
By mapping cognitive states natively within a pure 3D harmonic frequency topology (X, Y, Z space), we eliminate artificial high-dimensional bloat entirely. This is our own novel foundational research, changing the very base of how machine intelligence learns.
Operating strictly in 3D frequency space fundamentally solves the lifelong memory consistency problem. New knowledge and data can be added immediately in real time without disturbing, retraining, or destabilizing the underlying model.
NATSU — Sovereign 24M Cognitive Core
NATSU is Paras Lashkari's personal Jarvis-grade cognitive operating system, engineered to function as an autonomous digital brain orchestrating complex thought, coding environments, and system tools.
At its core sits a proprietary 24 Million Parameter compact foundation model, written and architected from scratch. Rather than chaining together third-party commercial APIs, NATSU operates with complete model self-sovereignty.
Engineered for edge-native execution with microsecond local latency, NATSU acts as the cognitive central nervous system — executing complex internal reasoning loops without leaking data or relying on remote cloud services.
System Peripherals
& Physical Loops
Intelligence cannot remain an isolated mathematical abstraction. We build sensory and operational loops around the model — granting it ears, eyes, and hands.
Acoustic Voice Resonance
Direct bidirectional auditory synthesis with low-latency conversational presence.
Spatial Vision Stream
Real-time scene geometry comprehension and symbolic interface parsing.
Physical & OS Automation
Deterministic execution loops connected directly to developer tools and physical controllers.
A Small Lab
With a Long View.
“I started Voidex because I wanted to understand intelligence by building it, not just using it.”
“We are a young, independent lab inside LGC. We work slowly, carefully, and from the ground up. Some of what we build will remain private until it is mathematically mature.”
Build it ourselves
We reject unexamined abstractions. Real machine sovereignty requires creating your models, memory layers, and weights from scratch.
Understand before scaling
Scale without clarity only magnifies inefficiency. We rigorously clarify cognitive mechanics in compact regimes before applying compute.
Ship what works
Zero speculative hype, zero synthetic benchmark marketing. What leaves the lab must demonstrate real, uncompromising stability.
Ventures Without Limits.
Voidex is the artificial intelligence and scientific research arm of Lashkari Group of Companies (LGC). As the group's computational innovation center, Voidex pioneers sovereign algorithmic research across our clinical, architectural, and software ecosystems.
Official academic inquiries, researcher dialogue, and collaboration requests.
Operating quietly out of Gujarat's commercial capital.
Non-commercial internal lab cycle. Inquiries reviewed on a rolling basis.