A tutorial describes building an autonomous data science agent using DeepAnalyze-8B. The workflow begins by preparing a stable runtime, installing machine-learning dependencies, and loading the DeepAnalyze tokenizer and model in 4-bit mode to keep the workflow practical on limited GPU memory. A sandboxed execution environment is then created, allowing the model to generate Python code, execute it safely, observe results, and continue analysis in an agentic loop. Finally, the agent is given a realistic multi-file e-commerce workspace to clean, join, analyze, visualize, and summarize the data as a structured analyst-grade report. The tutorial details installing dependencies including "transformers>=4.44", "accelerate>=0.30", "bitsandbytes>=0.43", "sentencepiece", "openpyxl", and pinning "numpy==2.0.2". The model ID is "RUC-DataLab/DeepAnalyze-8B", and it is loaded in 4-bit mode using BitsAndBytesConfig with parameters including "load_in_4bit=True", "bnb_4bit_quant_type='nf4'", and "bnb_4bit_compute_dtype=torch.float16". The model is loaded with "device_map='auto'" and "torch_dtype=torch.float16". After loading, the VRAM used is reported as "%.1f GB" % (torch.cuda.memory_allocated()