Maravilla Digital

Pioneering neuromorphic AI systems that think with spikes, not multiplications. Energy-efficient intelligence inspired by the brain.

Neuromorphic Computing Spiking Neural Networks On-Device AI

Our Mission

We're building brain-inspired AI models that combine Spiking Neural Networks (SNN) with State-Space Models (SSM) for temporal reasoning without degradation. Our goal: AI that runs efficiently on edge devices, consuming a fraction of the energy of traditional transformers.

Spiking Neural Networks

Leaky Integrate-and-Fire neurons with surrogate gradient training. Binary spike communication enables massive parallelism with minimal energy.

State-Space Memory

Mamba-inspired selective gating for long-range temporal dependencies. Deep reasoning without the quadratic cost of attention mechanisms.

Edge-First Design

Models optimized to run on Apple Silicon, mobile NPUs, and neuromorphic chips. AI that works offline, privately, on your device.

Research Focus

SpikeSSM: Hybrid Neuromorphic Architecture

Our core research combines the event-driven efficiency of spiking neurons with the temporal memory of state-space models:

Spike Encoding Layer — Transforms input into sparse binary spike trains (80%+ sparsity)
SSM Temporal Core — Selective state-space model captures long-range patterns without degradation
Spike Decoding Layer — Converts processed representations back to interpretable outputs
Sub-millisecond Inference — Achieving <1ms per sample on consumer hardware

<1ms

Inference Latency

80%+

Spike Sparsity

10x

Energy Efficiency Target

Why Neuromorphic?

Energy Crisis in AI

Large language models consume megawatts. The brain runs on 20 watts. Neuromorphic computing bridges this gap with event-driven, sparse computation.

Privacy by Design

Models small enough to run locally mean your data never leaves your device. No cloud dependency, no latency, no data exposure.

Temporal Intelligence

Real-world data is temporal — speech, video, sensor streams. Spiking networks process time natively, unlike transformers that tokenize it away.

Team

Based in La Serena, Chile. A small team with deep expertise in neural architecture design, compiler engineering, and on-device ML deployment. We believe the next breakthrough in AI won't come from bigger models — it will come from smarter architectures.