AI/ML Engineer / Generative AI Engineer**

ARCHIDIGITAL is looking for an AI/ML Engineer to develop interactive systems based on local neural network models.

We need a specialist who can independently deploy models on local GPU servers, combine language, voice, and video models into a single system, and optimize their performance for minimal latency.

Current Task

Interactive installation «Mishka's Soft Phone».

This is a children's phone with a microphone and speaker for a children's rehabilitation center, connected to a local server.

The user speaks with the character Mishka. The server must receive voice, recognize speech, form a response with a language model, synthesize it with the character's voice, and return audio to the phone with minimal latency.

It is necessary to develop the server-side of the system, including:

• audio stream reception and processing
• Speech-to-Text
• operation of the local language model
• dialogue logic and context
• Text-to-Speech and character voice imitation
• streaming data processing
• API for interaction with the device
• end-to-end latency optimization
• stable system operation on the local server

Next Direction of Work — GENMIRROR

GENMIRROR is an ARCHIDIGITAL interactive system for real-time neural network processing of video streams.

The specialist will participate in the further development of the project: connecting new generative models, building and optimizing local inference pipelines, working with image and video models, and developing the server architecture.

Main Requirements

• Proficient Python
• PyTorch
• Linux
• Docker
• CUDA and understanding of GPU operation
• experience deploying open-source models locally
• understanding of modern LLMs
• experience with Speech-to-Text and Text-to-Speech
• experience with voice cloning or voice conversion
• understanding of streaming inference
• ability to build systems from several interconnected AI models
• experience optimizing inference and latency
• understanding VRAM, quantization, and GPU resource management
• ability to independently study, launch, and integrate new open-source models

Will be an Advantage

• experience with Whisper, faster-whisper or equivalents
• experience with vLLM, llama.cpp, TensorRT, ONNX Runtime
• experience with modern TTS and voice cloning models
• experience with Stable Diffusion, FLUX or generative video models
• LoRA and fine-tuning
• WebSocket, gRPC or WebRTC
• experience developing real-time AI systems

Who We Are Looking For

We need an engineer who can work not only with external APIs but directly with models, local infrastructure, and GPU inference.

It is important to be able to independently select a model, deploy it, integrate it into a system, optimize it, and bring it to stable operation.

Work Format

Project-based work with the possibility of long-term collaboration.

In your application, please include your GitHub or project examples and briefly describe your experience with local LLMs, STT, TTS, voice cloning, and GPU inference.

Candidates from St. Petersburg and Leningrad Oblast are considered first due to the need for in-person meetings for projects.