Intelibot Scribe

Intelibot Scribe drives the complete scientific lifecycle: hypothesis formulation, firewalled literature retrieval, sandboxed model training, iterative hyperparameter retuning, publication-grade LaTeX drafting and live plagiarism verification. Literature retrieval queries Crossref and open academic APIs without exposing the agent to untrusted instruction injection, and all machine learning code runs inside isolated E2B container sandboxes on CPU-only PyTorch. A two-step data-driven selection stage searches Kaggle for the most relevant dataset, cleans and normalises features through a ten-step preprocessor, and picks between RandomForest, GradientBoosting, PyTorch and MLP models based on dataset characteristics. Users can upload one or two reference PDFs so the writing agent mirrors their cadence, vocabulary and tone. Drafts are scanned against GoWinston AI's corpus with an interactive originality gauge, and a three-tier model fallback keeps the pipeline immune to rate limits.