RICLAB

Independent laboratory · Santiago, Chile

Advanced science,
within everyone's reach.

We research, build, and share.We make scientific knowledge accessible—and create new knowledge—across artificial intelligence, autonomous systems, and advanced computing.

Research & development

Projects built to move intelligence into the real world.

Four active lines of research spanning embodied systems, data infrastructure, private edge intelligence, and digital trust.

04 / Active projects
01

Biologically inspired embodied and distributed AI

Autonomous Digital Nervous System

An artificial intelligence architecture that emulates the human nervous system. It moves beyond traditional centralized models through a hierarchy of hyper-specialized micro-models that interact with the physical world in real time.

Research focus

  • Dual architectureUltra-fast peripheral reflexes at the edge (System 1), backed by a central model for complex reasoning (System 2).
  • Latent communicationText-free internal transmission through continuous tensors enables physical responses in milliseconds.
  • Closed-loop learningProgressive physical training in simulation for autonomous motor and sensory control.

Key technologies

  • Simple Attention Networks
  • Spiking Neural Networks
  • Latent Vector Bus
02

Neural text embeddings generated directly inside PostgreSQL

pg_embedder

A PostgreSQL extension written in Rust that generates semantic feature vectors without external services or runtime model files. Model weights are embedded directly into the compiled binary.

View repository

Research focus

  • Self-contained deploymentModels and tokenizers are compiled into the library with the Candle machine-learning stack.
  • Dual storageWorks natively with FLOAT4[] arrays or with pgvector and its HNSW/IVFFlat nearest-neighbor indexes.
  • Broad compatibilitySupports PostgreSQL 13 through 18 through pgrx feature management.

Key technologies

  • Rust
  • pgrx
  • Candle ML
  • PostgreSQL
  • pgvector
03

Private artificial intelligence for everyday hardware

Ultra-light On-Device AI

A new generation of AI models designed to run locally on any device, from phones to wearables. Native 1-bit and 2-bit training removes cloud dependency for fast, efficient, and fully confidential personal assistants.

Research focus

  • Native speedComplex computation becomes highly efficient operations for affordable hardware.
  • Absolute privacyExecution stays on the device; personal and usage data never leaves it.
  • Minimal consumptionLow memory requirements and a reduced impact on battery life.
  • Local adaptabilityContinuous on-device learning and retraining adapts to each user.

Key technologies

  • 1-bit & 2-bit precision
  • Edge AI
  • On-device learning
  • Local inference
04

Advanced research against synthetic media and identity impersonation

AI Labs: Generative Fraud Prevention

A research program that anticipates emerging attack vectors by studying and reproducing generative fraud techniques, then turning those findings into reliable identity defenses without compromising user conversion.

Visit Idesify

Research focus

  • Liveness detectionBlocks presentation attacks such as masks and photos, plus injection attacks from virtual cameras.
  • Deepfake defenseIdentifies synthetic faces and subtle generative artifacts in spatial and frequency domains.
  • Document forensicsPixel-level detection of tampering, copy-move edits, and AI-manipulated fields across 150+ identity document types.
  • Regulatory complianceLegal NLP and RAG models grounded in GDPR and KYC/AML texts automate compliance without hallucinations.

Key technologies

  • Liveness detection
  • Spatial & frequency analysis
  • Pixel-level forensics
  • Legal RAG