The secondary use of clinical research data for scientific publication remains a significant bottleneck in medical research, as collected datasets are rarely transformed into disseminated findings due to the complexity of the data and the writing process. In this paper, we propose an open-source three-dimensional personalization framework (Lab, Personal, Global), to dynamically control these knowledge influences in the paper preparation. The framework extracts ML-based writing style features from reference papers and constructs adaptive prompts based on the set influence levels. The system supports four AI vendors (Groq Llama 3.3 70B, Google Gemini, OpenAI GPT-4, and local GPT-OSS 120B) through a unified abstraction layer. An evaluation with 20 researchers showed significant improvements in writing consistency over baseline methods, along with strong usability scores (SUS 80.78). The framework was validated using real-world clinical data from the RAISE platform, demonstrating its potential to accelerate the transformation of research datasets into publishable scientific output.