Answers are grounded in this paper and cite their sources.
Large Language Models (LLMs) can write code, debug programs and draft documentation, but the quality of their output depends heavily on the prompt, and most non-technical users cannot write specific, constrained and contextrich prompts.This paper presents SmartPrompt (Smart Prompts, Smarter Language), an automated prompt-engineering web platform that sits between the user and the LLM.A rough idea, entered as text, speech or an image, passes through a four-stage meta-prompting pipeline (Discovery, Compose, Critique and Evaluate) that returns three optimized prompt variants, each guided by a different technique from a taxonomy of 18 prompt-engineering techniques and scored on a six-axis rubric.The chosen prompt can be run concurrently on several LLM providers to compare answers and latency side by side.The system learns from use through a scikit-learn intent classifier (HashingVectorizer with SGDClassifier, updated by partial_fit), a Beta-posterior technique selector, rubric-weight regression on user ratings and dynamic fewshot retrieval of user-edited prompts.It is built with React, Vite and Tailwind CSS on the front end and FastAPI with SQLite on the back end, and streams stage progress using Server-Sent Events.In testing, the local learning components added well under 5 ms per request, intent accuracy on 45 held-out requests rose from 46.7% to about 60% with 18-36 labelled examples, and all 22 API cases and 15 automated tests passed.
No takes yet. Share an insight, caveat, or question.
P et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: