Backend Developer for Voice Technology Team

Backend Developer for Voice Technology Team

We develop high-load speech recognition (ASR) and speech synthesis (TTS) services that power Yandex's Alice, Browser, Translator, and other products, used daily by millions of people.

Our team is responsible for the backend infrastructure of these technologies: from designing and developing gRPC services to optimizing inference for modern neural network models.

Voice technologies are currently undergoing rapid development: just a few seconds of your voice can be used to synthesize any phrase or even make a model sing a song in your style. But behind these capabilities lie increasingly complex models, often combined in cascades with non-trivial inference logic. This poses new challenges for us: how to maintain high performance, scalability, and stability when latency requirements are measured in milliseconds and load is measured in thousands of requests per second.

We expect you to

Implement new speech synthesis and recognition models You will work closely with ML teams, understand the architecture of new models (TTS, ASR), design efficient inference schemes, and adapt our services to their specifics — considering strict requirements for latency, throughput, and stability.

Develop high-load gRPC services from scratch You will write high-performance, testable, and fault-tolerant code in C++ for new features and services that will later be integrated into Alice, Search, Translator, and other products.

Optimize neural network inference You will research and implement modern inference engines (vLLM, SGLang, TensorRT-LLM), experiment with batching, quantization, and caching — anything that helps accelerate models without losing quality.

Improve service reliability You will participate in the full development cycle — from design and testing to deployment and support. An important part of our work is improving monitoring, adding metrics and logs, and automating release processes.

More about backend at Yandex — in the channel Yandex for Backend

Will be a plus

• Familiar with modern frameworks for LLM model inference: SGLang, vLLM, TensorRT-LLM
• Have experience with NVIDIA GPUs: understand GPU architecture, have developed or optimized algorithms using CUDA or Triton

Work Requirements

• Proficient in C++, preferably C++17 and above
• Have developed high-load backend services: able to design systems with predictable latency, stable throughput, and to find and eliminate memory leaks and performance degradation under load
• Have basic knowledge in the field of ML and are familiar with PyTorch — our team does not train models, but it is important to understand how inference works and sometimes be able to read ML code