Job Details

HaaS Platform is looking for an AI Engineer (Agents) Middle / Senior to develop a platform for virtual AI agents. Salary from $4500. Location: Tashkent (until September), then remote. Full-time employment.

We are a product team building a platform focused on virtual AI agents for communication automation. We work fast, without excessive bureaucracy, experiment a lot, and implement new approaches. We are currently strengthening our agent systems direction and are looking for an engineer who understands how it works not only from the outside but also "under the hood". Your responsibilities will include: - Development and enhancement of voice and text AI agents (pipeline: STT → LLM → TTS) - Building and improving agent scenarios: prompting, tool use, orchestration - Optimizing latency, response quality, and pipeline stability - Fine-tuning and evaluating LLM and speech models for real-world use cases - Integrating AI components into production (C# / SIP / RTP stack) - Researching and implementing new models, frameworks, and approaches Our tech stack: - Python, C# - Whisper (STT), Qwen / LLaMA / Mistral (LLM), Kokoro (TTS) - LangChain / LlamaIndex - vLLM / Ollama / llama.cpp - SIP / RTP, Windows / Linux

We expect: - 2+ years of experience in AI/ML in production - Understanding of LLMs: prompting, RAG, fine-tuning - Experience with agent systems - Understanding of how models and pipelines work - Experience with speech (STT / TTS) - will be a strong plus - Ability to evaluate model quality end-to-end - Experience with on-premise inference (local deployment, quantization, latency) - Confident work with Linux and/or Windows What's important: - Interest in agent systems and AI - Readiness to work in a dynamic environment: many varied tasks - Ability to independently figure things out and bring solutions to production Will be a plus: - HuggingFace ecosystem - Experience in Telecom / VoIP Agentic AI (tool use, multi-agent systems)

Work format: - Relocation to Tashkent for the launch period (until September) - Afterwards - full remote - Flexible cooperation format