Based in Almaty, Kazakhstan.
Building Manifestro / Sanday.
Bagzhan Karl is a researcher and engineer focused on the intersection of biology and neural architecture. By studying the principles of living systems, he develops architectures that mimic the fluid adaptation and temporal complexity of nature.
Bagzhan began his work in technical systems and project development in 2020. That same year, he enrolled in the International IT University (IITU) in Almaty to study Information Systems (IS). In 2022, he decided to leave his formal studies to focus on full-time research and production-grade engineering in the field of deep learning.
Prior to founding Manifestro, Bagzhan focused on implementing and managing complex integrations with existing LLM APIs. During this period, he identified major limits in centralized, cloud-based AI services: latency, opacity, and weak temporal adaptation. Those constraints led to a different direction—building transparent, decentralized systems with stronger continuous-time behavior.
This technical requirement became the catalyst for the Manifestro project. Bagzhan shifted focus from building on top of external APIs to developing proprietary architectures from the ground up, rooted in biological dynamics. He currently leads the development of sanday, utilizing nature-inspired models as an alternative to standard Transformer architectures. In Phase 2 of the project, the system achieved significant gains in neural fidelity, proving the feasibility of high-performance ASR and TTS through biological mimicry.
Focus Areas
Biological Neural Networks, Nature-Inspired AI, Continuous-Time Dynamics, Neural Fidelity.
Current Milestone
Phase 2 Completion (Neural Fidelity Milestone).
Bagzhan’s approach to artificial intelligence is grounded in biological signals such as synaptic pruning and adaptive state evolution. Applied to neural networks, this means reducing redundant pathways while preserving the dynamic structures that support temporal reasoning. The goal is to isolate functional intelligence that behaves closer to living cognition.
The project prioritizes empirical results over industry trends. Progress is measured through verifiable benchmarks, behavioral quality, and reproducible experiments. The objective is to expand what LNN architectures can represent in real-world speech systems.
The long-term roadmap for Manifestro (Phases 6–8) includes public API access for speech recognition and synthesis. As the architecture matures, deployment cost and hardware requirements are expected to improve naturally. The project serves as a practical implementation of adaptive, accessible neural architecture.