AI Prompt Studio: A Unified Tool For Prompt Design, Comparison And Feedback

AI Prompt Studio Semester 2 delivers a concrete and measurable improvement over its predecessor by replacing expensive multi-model parallel execution with a domain-aware single-model routing strategy backed by persistent MongoDB memory. The hybrid learning approach — combining user-guided domain initialization for first-time domains with fully automatic memory-based routing for all subsequent prompts — achieves a 66% reduction in API calls, approximately 67% reduction in response latency, and marginal improvement in GPT-4 evaluation scores across Relevance, Creativity, and Factual Accuracy dimensions. The system demonstrates that domain awareness, implemented through a simple key-value memory store and a lightweight routing decision in the FastAPI backend, is sufficient to deliver substantial engineering value without requiring complex machine learning infrastructure. The modular architecture supports straightforward addition of new models or domains. The core principle established by this work — learn the domain-model relationship once, route intelligently thereafter — represents a broadly applicable design pattern for cost-effective, quality-preserving LLM system design in real-world deployments.

[1] L. A. Loss and P. Dhuvad, "From Manual to Automated Prompt Engineering: Evolving LLM Prompts with Genetic Algorithms," 2025 IEEE Congress on Evolutionary Computation (CEC), pp. 1–8, 2025. DOI: 10.1109/CEC65147.2025.11043059 [2] R. Deshmukh, S. Gurav, R. Raut, Y. Patil, and M. Bhavsar, "Optimizing Human-AI Interaction: Innovations in Prompt Engineering," in Proc. 3rd International Conference on Intelligent Data Communication Technologies and Internet of Things (IDCIoT-2025), IEEE, pp. 1240–1246, 2025. DOI: 10.1109/IDCIOT64235.2025.10914815 [3] W. Li, X. Wang, W. Li, and B. Jin, "A Survey of Automatic Prompt Engineering: An Optimization Perspective," arXiv preprint arXiv:2502.11560, 2025.