Data Science & AI Models Prompts

A retrieval-augmented system prompt that answers only from context and returns inline citations or 'I don't know'.

Produce a reproducible Pandas cleaning pipeline: types, missing values, dedupe, outliers.

Gives you a reproducible, well-structured PyTorch training script — config, model, dataloaders, train/eval loop, metrics, checkpointing, and early stopping — tuned to your task.

Produces a hardware-aware Ollama model recommendation for your task plus a ready-to-run install and run script with VRAM checks, instead of guessing a model name and hoping it fits.

Produces a complete, local-first RAG pipeline with document loading, chunking, Ollama embeddings, a vector store, retrieval, and a grounded answer step with citations, requiring no cloud APIs.

Produces a privacy-first local chat configuration with model and quantization choice for your hardware, a system prompt, conversation settings, and a data-leakage audit checklist so nothing leaves your machine.

Produces a custom PyTorch Dataset with correct transforms, a tuned DataLoader, and a debuggable augmentation pipeline that handles edge cases instead of throwing on the first weird sample.

Produces a transfer-learning script that swaps the right head, freezes the right layers, and uses distinct learning rates so you adapt a backbone instead of nuking its pretrained weights.

Produces a reusable schema-validation and dtype-coercion script that flags every column that drifted from spec, coerces what it safely can, and quarantines what it cannot instead of producing silent NaNs.