Vivum AI, founded in 2021, develops efficient neural network solutions for autonomous systems operating in resource-constrained environments.

The company's core technology uses dynamic neural networks that require significantly fewer computational resources than traditional AI approaches. Their solution addresses the challenge of implementing autonomous systems in environments where power, space, and computational resources are limited, such as downhole drilling operations and GPS-denied navigation.

The technology implements bio-inspired AI using differential equations with embedded time constants, allowing for efficient processing of time-series data. The system can be deployed on edge computing devices like FPGAs and microcontrollers, enabling real-time autonomous decision-making without requiring powerful GPUs or cloud connectivity.

The company's proposed solution for the oil and gas industry is a holistic system that can control not only the rotary steerable but also all the surface equipment, optimizing parameters like temperature, pressure, shock, vibration, RPM, weight on bit, and direction. The goal is to deploy this AI-powered system downhole in the rotary steerable to provide real-time control signals and optimize the drilling process.

Edge-deployed AI software solution (with optional hardware) + subscription-based licensing model for autonomous drilling control systems, but they are still in the pre-proof of concept stage for oil and gas.

Vivum AI's platform leverages innovative neural network architecture specifically designed for edge computing and constrained environments:

• Dynamic Neural Networks

  • Uses differential equations with time constants versus traditional matrix multiplication
  • Achieves same functionality with 95% fewer neurons (19 vs 400 in documented test case)
  • Optimized for time series data processing
  • Enables real-time decision making on edge devices

• Bio-inspired Architecture

  • Based on mapped responses from biological neurons
  • Employs evolutionary algorithms for training versus gradient descent
  • Designed specifically for signal/power/space constrained environments
  • Provides explainable AI through smaller network size and decision tree integration

• System Integration

  • Compatible with existing hardware systems or can provide hardware solution
  • Supports API integration for data exchange
  • Includes cloud infrastructure for storage and computing
  • Modular design allows embedding in various applications

• Designed for downhole deployment on FPGA/microcontroller hardware
• Real-time integration of downhole sensor data (temperature, pressure, shock/vibration)
• Control capabilities for RSS direction, WOB, RPM, and mud flow parameters
• API integration ready for rig system connectivity
• Reduced computational footprint (19 vs 400 neurons) compared to traditional neural networks