// the find
promptslab/Promptify
Prompt Engineering | Prompt Versioning | Use GPT or other prompt based models to get structured output. Join our discord for Prompt-Engineering, LLMs and other latest research
Promptify wraps LLM calls into scikit-learn-style task objects (NER, Classify, QA, etc.) with Pydantic output schemas, using LiteLLM as the backend so you can swap between OpenAI, Anthropic, and local models by changing a string. It's aimed at developers who want structured NLP outputs without writing prompt plumbing from scratch. The 3-line API is genuinely clean for standard tasks.
LiteLLM backend means provider switching is real and works — not just a theoretical abstraction. Pydantic output schemas give you type-safe results instead of hoping the LLM returned parseable JSON. The task taxonomy (NER, Classify, QA, SQL, Relation Extraction) covers the boring-but-common NLP cases that most teams end up reinventing. Built-in cost tracking via get_cost_summary() is a thoughtful addition that most similar libraries skip.
The repo has 4.6k stars but the directory tree shows no CHANGELOG and the last meaningful activity looks thin — hard to tell if the current README reflects the actual installed package or aspirational docs. Domain specialization via domain='medical' is a fuzzy concept with no clear mechanism shown; it likely just injects the word into the prompt, which is a weak abstraction. The fallback JSON parser is a code smell — if your provider doesn't support structured outputs natively, you're relying on string-parsing heuristics that will fail on edge cases. No streaming support visible, which matters for long summarization or SQL generation tasks where you want incremental output.