Prompt Enhancer

Inspiration

• Observing how non-technical users struggle to get high-quality outputs from LLMs due to a lack of "prompt engineering" knowledge.
• The goal of creating a "middleman" that optimizes cost and time by reducing the need for repetitive "try-again" prompting cycles.

What it does

• Injects expert-level prompt engineering techniques into basic requests using a Retrieval-Augmented Generation (RAG) architecture.
• Allows users to select specific personas and reasoning modes through a Chrome Extension UI that talks to a FastAPI backend.
• Calculates and displays real-time metrics, such as estimated tokens and minutes saved.

How I built it

• Developed a FastAPI backend that utilizes ChromaDB as a vectorDB to retrieve prompt engineering research on the fly.
  • Integrated the Gemini model via LangChain to synthesize the retrieved context into a final, optimized prompt. Built the frontend as a Chrome Extension using vanilla JavaScript and CSS for a lightweight, responsive user experience.

Challenges we ran into

• Fine-tuning the RAG retrieval window to ensure the context was helpful without exceeding the LLM's prompt length or distracting the model.

Accomplishments that we're proud of

• Successfully implementing a full RAG pipeline that runs locally.
• Developing a persistent metrics dashboard that tracks user impact across sessions using browser local storage.
• Creating a "copy-to-clipboard" workflow that fits naturally into a user's existing AI-assistant habits

What we learned

Too Much!

What's next for Prompt Enhance

• Porting the backend to a serverless architecture to provide a truly "always-on" global service for a wider user base
• Allow the user to personalize the choice of the RAG documents.

Built With

• chromedb
• css3
• fastapi
• html5
• huggingface
• javascript
• langchain