Evgeny Trifonov

The application specializes in creating engaging and educational mathematical games for schools, classrooms, and individual students.

Focusing on providing seamless real-time fixes and upgrades, I excel in developing project architectures, microservices, back-end and front-end systems, database structures, network protocols, and high-load optimized server infrastructures. Additionally, I provide valuable guidance to teams, conduct thorough code reviews, create CI/CD pipelines, end-to-end and high-load testing frameworks, and produce comprehensive technical documentation. My expertise lies in identifying and addressing project weaknesses to drive exponential growth, resulting in a significant four-fold increase in real-time online application usage.

ACHIEVEMENTS
• Successfully identified and addressed project weak points, resulting in a significant four-fold increase in real-time online application usage.
• Designed and implemented high-load optimized server infrastructure, ensuring reliable and scalable performance.
• Developed end-to-end testing frameworks that improved code quality and reduced potential issues.
• Created comprehensive technical documentation, facilitating seamless knowledge transfer and project maintenance.

BluepJS, as a Blueprint Visual Scripting, is a no-code scripting system using a node-based interface. This system is highly flexible and powerful as it can use the full range of concepts and tools generally only available to programmers. BluepJS is inspired by the Unreal Engine Blueprint system and is designed as a scripting engine for other projects.

Takatan JavaScript is open-source software you can use to create chatbots, websites, and communication between these and other user interfaces. I was the architect and developer of the project and built it with a real-time reactive core to support websites and chatbots with visually appealing UX.

I was the full-stack developer and server administrator for the project. I acted as a kind of IT/technical mage for the team and supported their ideas from the technical side. I ensured website security and regular updates and oversaw mobile app development, video streaming, and other functions. To this day, I am on excellent terms with the Phanganist team and coach the founder in programming the project.

This project focuses on converting a Qwen3-TTS model from the SafeTensors format into the ONNX representation to enable efficient inference using C++ and ONNX Runtime. By leveraging ONNX, the model can be executed without a Python runtime, making it ideal for integration into native applications and real-time systems.

KEY GOALS
• Convert a Qwen3-TTS SafeTensors model to ONNX
• Enable inference via C++ using ONNX Runtime
• Optimize the graph for performance and memory efficiency
• Support deployment in low-latency environments

TECHNICAL SCOPE
The project involves:

• Loading and interpreting SafeTensors weights
• Reconstructing the model graph
• Exporting the model to ONNX
• Handling dynamic axes (text length/audio frames)
• Addressing TTS-specific components (encoder/decoder/vocoder)

OUTCOME
The result is an ONNX-compatible Qwen3-TTS model that can be directly executed within a C++ application via ONNX Runtime, enabling:

• Python-free deployment
• Faster startup time
• Reduced runtime dependencies
• Improved portability

USE CASES
• Embedded AI applications
• Real-time speech synthesis
• Desktop/native inference tools
• Cross-platform TTS engines