Applied Scientist (LLM)

SQUAD is a distributed team focused on developing high-performance Generative AI features across Cloud and Edge environments.

• 3+ years of commercial experience in Machine Learning, focusing on NLP or LLM
• Strong knowledge of Python3, NumPy, pandas, PyTorch, Hugging Face (Transformers, PEFT, Accelerate)
• Proficiency in PEFT/LoRA and Reinforcement Learning techniques
• Deep understanding of attention mechanisms, tokenization, context window management, and embedding spaces
• Practical experience with Retrieval-Augmented Generation (RAG), Fine-tuning, or Agentic frameworks
• Ability to manage and analyze large datasets (>100GB) across text, image, and audio
• Experience building high-fidelity datasets and robust data pipelines
• Expertise in prompt engineering, agentic framework design, and LLM pipeline orchestration
• Experience deploying LLMs using Triton Inference Server, vLLM, TGI, or ONNX
• Good written and spoken English

Nice to have:

• Experience with Pinecone, Weaviate, Milvus, or Chroma
• Advanced quantization (GGUF, AWQ, EXL2), pruning, and knowledge distillation
• Experience with LangChain, LlamaIndex, or AutoGen
• Basic understanding of web/client-server architecture and streaming API responses (Asyncio, aiohttp)
• Familiarity with RAGAS, DeepEval, or G-Eval
• Experience using Docker, Kubernetes, and cloud GPU orchestration (Run:ai, Lambda Labs)
• Knowledge of C++, Triton, or CUDA for custom kernel development

• Design and implement advanced methods in prompt orchestration, fine-tuning (SFT/RLHF/DPO), and autonomous agentic workflows
• Curate high-quality training data from large-scale text and multi-modal sources
• Identify patterns in model hallucinations and visualize evaluation metrics
• Tune hyperparameters and improve inference speed/accuracy through PEFT (LoRA/QLoRA) and advanced prompt engineering
• Collaborate with Product and Data Engineering teams to integrate LLM features
• Track and report progress using benchmarks (MMLU, HumanEval) and internal KPIs
• Stay updated on new techniques (e.g., State Space Models, Transformer variants) and evaluate for production readiness
• Mentor junior colleagues to enhance team expertise

Please be informed that the data controller is SQUAD POLAND Sp. z o.o. ("controller"). You have rights regarding your personal data including access, rectification, erasure, restriction, objection, portability, and complaint to supervisory authority. Data processing is for recruitment purposes. Provision of mandatory data is required by law; refusal may prevent recruitment. Data processed until recruitment completion or consent withdrawal. Data recipients include Just Join IT and other entities involved in recruitment processing.