AI Engineer

Responsibilities

Design, build and maintain LLM-based applications such as chatbots, assistants and agents.
Fine-tune and adapt open-source language, speech and vision models to the Mongolian language and to specific domains.
Develop and improve our speech, document and vision products, such as Mongolian speech-to-text (STT), text-to-speech (TTS), OCR and eKYC, across the full lifecycle, from data preparation and training to deployment and monitoring.
Package models as reliable, low-latency APIs and services, and integrate them into our products.
Build evaluation datasets and metrics (accuracy, WER/CER, latency, hallucination rate) to measure model quality and track improvements over time.
Follow new models and techniques, run focused experiments, and bring the approaches that work into our products.

Comfortable writing code in Python.
Familiarity with LLMs, prompt engineering, RAG, embeddings and vector databases.
Have used PyTorch and libraries such as Hugging Face Transformers.
Familiarity with REST APIs (for example FastAPI) and Docker.
Eager to learn: you read research papers and like turning ideas into prototypes.
A bachelor's degree or higher in Computer Science, Artificial Intelligence, Mathematics or a related field is an advantage.

Experience fine-tuning models with LoRA, PEFT or similar methods.
Speech (ASR/TTS) or natural language processing (NLP) projects, especially with Mongolian or other low-resource languages.
Experience serving models efficiently with tools such as vLLM, Triton Inference Server or ONNX Runtime.
Experience building LLM agents that use tools and external systems.