Bitmanager.ai is an AI startup operating in one of the fastest-growing areas: conversational AI agents.
We build a platform for voice and text AI agents used in sales, customer support, consulting, and debt collection. Our agents conduct conversations, work with customer data, call tools tools, and interact with enterprise systems.
We are looking for a strong Speech Engineer who will own either ASR or TTS, develop multilingual voice capabilities, and ensure that speech services perform reliably in production.
The goal is clear: an agent that understands the speaker, responds quickly, and sounds natural across all supported languages.
Responsibilities
- Own and develop either streaming speech recognition (ASR) or speech synthesis (TTS), including support for low-resource languages.
- Measure and reduce end-to-end latency from the end of the user’s utterance to the first audio response.
- Improve real-time conversation quality: endpoint detection, pauses, interruptions, and natural turn-taking.
- Handle telephony-specific challenges such as narrowband audio, streaming, echo, packet loss, and real call recordings.
- Fine-tune and adapt speech models together with data and language specialists.
- Build evaluation systems combining offline metrics, end-to-end call testing, and regression checks.
- Own the production quality of GPU-based speech services, including deployment, monitoring, incident investigation, and root-cause analysis.
- Define technical direction, review code, support other engineers, and contribute to on-premise customer deployments.
Must-have qualifications
- Strong Python skills and experience building production services, not only research scripts.
- Hands-on experience with ASR and/or TTS, including model integration, deployment, and production operation. Research-only or model-training-only experience is not sufficient.
- Understanding of streaming audio processing, chunking, buffering, latency sources, and real-time constraints.
- Experience with PyTorch, GPU inference, and performance optimization.
- Ability to define quality metrics, build evaluation datasets, and distinguish meaningful improvements from noise.
- Production experience with Docker, logging, monitoring, and debugging issues that appear only in live systems.
- Ability to independently own a technical direction and make decisions with incomplete information.
- Strong communication skills and the ability to work with product teams, linguists, infrastructure engineers, and customer technical teams.
- Professional proficiency in Russian and English.
- Availability to work regularly from our Moscow office. Fully remote work may be considered in exceptional cases for candidates within a time zone close to Moscow.
Experience with voice agents is not required. Strong speech engineering fundamentals, production maturity, and the ability to understand and improve an existing pipeline are more important.
Nice to have
- Telephony experience: SIP, RTP, codecs, echo cancellation, or 8 kHz audio.
- Experience building real-time voice agents or dialogue systems.
- Experience with VAD, endpoint detection, and interruption handling.
- Experience fine-tuning TTS models using SFT or LoRA, or working with reference-based voice cloning.
- Experience with multilingual speech models and low-resource languages.
- Inference optimization experience: batching, quantization, ONNX, TensorRT, or vLLM.
- Experience deploying ML services on-premise in enterprise environments.
- Experience integrating LLMs into real-time voice pipelines.
If you want to own how an AI agent hears, responds, and sounds in real conversations - not just improve an offline metric - we would be happy to meet you.
