Machine learning and AI engineering grounded in real implementation — from edge inference to production ML systems.
At BolivarTech, AI is not a buzzword — it is an engineering discipline. We design and build machine learning systems that solve real problems: computer vision running on embedded hardware, reinforcement learning agents for automated decision-making, signal processing pipelines for noisy sensor data, and ML models deployed reliably in production environments.
Our work spans the full ML lifecycle — from data exploration and model architecture design through training, validation, optimization, and deployment — with particular expertise in edge AI, where AI meets constrained hardware and the physical world. A key differentiator: we implement ML systems in Rust alongside Python, bringing memory safety, deterministic performance, and zero-cost abstractions to production AI pipelines and embedded inference runtimes.
AI and ML disciplines we research, build, and deploy.
Deploying machine learning models on microcontrollers and SBCs — STM32, NXP i.MX, Raspberry Pi, and custom hardware. Model quantization, pruning, and optimization for real-time inference using OpenVINO, TensorFlow Lite, Hailo-8, and bare-metal Rust runtimes running directly on the target hardware without OS.
Object detection, tracking, counting, and image restoration systems. Production experience with YOLO-based architectures, Vision Transformers (ViT), and codebook-based restoration models for edge and server deployments.
RL agent design and training for automated decision-making systems — Proximal Policy Optimization (PPO), genetic algorithms, and hybrid RL+evolutionary approaches. Production RL systems implemented in Rust using the Burn framework for high-throughput, low-latency inference loops.
ML-enhanced signal processing for embedded systems — Kalman filter tuning with evolutionary algorithms, rolling statistics, polynomial approximation, GNSS/NMEA data processing, and motor control signal analysis.
Production ML pipeline design — data ingestion, feature engineering, model training, versioning, evaluation, and deployment. High-performance data pipelines written in Rust for throughput-critical workloads. CI/CD for ML models, experiment tracking, and monitoring for model drift in production.
Dynamic ML-based risk assessment systems, statistical anomaly detection for time-series sensor data, and predictive models for industrial and financial applications with explainability requirements.
Frameworks, tools, and platforms we work with.
Representative AI systems and projects we design and deliver.
Real-time object detection, person counting, and trajectory tracking on embedded hardware — production experience with YOLO models on OpenVINO-optimized pipelines and dedicated AI accelerators (Hailo-8, Coral Edge TPU). No cloud dependency, privacy-preserving, low latency.
RL-based trading agents using PPO and genetic algorithms, with a Temporal Predictive Coding (TPC) world model encoder for latent state representation under partial observability — built entirely in Rust for performance.
ML-driven analysis of motor current sampling data, PWM duty cycle fitting, and evolutionary Kalman filter tuning for noise-robust signal processing in embedded control systems — running on STM32 and NXP targets in Rust.
GNSS multi-system NMEA parsing, multilateration algorithms for asset tracking, and ML-enhanced positioning for IoT and industrial location systems.
Applied research integrating Predictive Coding (PC) and Temporal Predictive Coding (TPC) as world model encoders in RL systems. TPC builds robust latent representations that capture temporal dependencies — currently a frontier research area with minimal production adoption industry-wide, which we are actively bridging toward real applications.
Whether you need AI running on constrained hardware, a production ML pipeline, or expert guidance on model architecture, we can help. Tell us about your problem and data.
Contact Us Back