حلول فيلينتس بالذكاء الاصطناعي تخدم أكثر من 5 وزارات حكومية في المملكة العربية السعودية والكثير من الشركات الكبيرة والناشئة في مختلف القطاعات مثل مجالات التوظيف، التعليم، تطوير الموظفين، تأهيل الطلاب، خدمة العملاء، المبيعات، وغيرها. وذلك من خلال تطوير واجهات برمجة التطبيقات مثل المحادثات التفاعلية مع الذكاء الاصطناعي، تحويل الصوت والفيديو إلى نص، .تحليل الفيديو، قراءة ومطابقة المستندات، اختبارات المهارات، ونظام مكافحة الغش. والمزيد والمزيد.. تعرف أكثر على حلول فيلينتس بالذكاء الاصطناعي من خلال موقعناAbout the role We’re looking for a mid-level AI Engineer who lives close to the model. Your core work is training and fine-tuning foundation models across language and speech — large language models (LLMs), automatic speech recognition (ASR), and text-to-speech (TTS). Beyond training, you’ll also build GenAI applications, integrate models into agentic workflows, and own the MLOps that keeps everything running in production. It’s a full-stack AI role with model training at its heart. Core responsibilities — model training & fine-tuning Fine-tune LLMs using parameter-efficient methods (LoRA, QLoRA, PEFT, adapters) and full fine-tuning where needed. Train and fine-tune ASR models for custom vocabularies, accents, and low-resource languages. Develop and fine-tune TTS systems for natural prosody, voice cloning, and multilingual synthesis. Curate, clean, and prepare high-quality datasets — text corpora, audio transcriptions, and speech recordings. Run rigorous evaluation: task metrics for LLMs, WER/CER for ASR, MOS and naturalness for TTS. Apply alignment techniques (RLHF, DPO, SFT) to steer LLM behavior toward desired outcomes. Supporting responsibilities — GenAI, agents & production Build LLM-powered applications including RAG pipelines, prompt chains, and conversational systems. Integrate trained models into agentic workflows with tool-calling, memory, and orchestration (LangChain, LangGraph). Optimize models for inference — quantization (INT4/INT8/GPTQ), pruning, distillation, and ONNX/TensorRT export. Familiarity with vllm, sglang, trt Own MLOps: CI/CD for models, containerized deployment, versioning, and production monitoring. Track model drift, latency, cost, and accuracy in production; respond to incidents and retrain as needed. Collaborate with product and engineering to turn business needs into shipped AI capabilities. Requirements MUST HAVE 2–5 years hands-on training or fine-tuning deep learning models. Strong Python and PyTorch; comfortable modifying model source code. Deep experience with the HuggingFace ecosystem (Transformers, Datasets, PEFT, Accelerate). Solid grasp of transformer architectures and tokenization. Audio processing with torchaudio, librosa, soundfile. Experience deploying and monitoring models in cloud environments. NICE TO HAVE Low-resource or multilingual speech/language model training. Distributed training experience (DeepSpeed, FSDP, or Megatron-LM). Experience building agentic systems end-to-end. Speech data collection, annotation, or augmentation pipelines. Background in phonetics, linguistics, or acoustic modeling. Tools & stack Python, PyTorch, HuggingFace, DeepSpeed, FSDP, Weights & Biases, MLflow, ONNX / TensorRT,Vllm,Sglang,langchain,langgraph, torchaudio, , Docker, SLURM / Kubernetes, AWS / GCP, Git + DVC. Education Bachelor’s or Master’s in Computer Science, Electrical Engineering, Computational Linguistics, or a related field. A strong portfolio of model training work is equally valued.