PromptBridge

Turns a rough, vague prompt into a clear one that gets better answers from an LLM.

Most people type a half-formed prompt and get a half-useful answer. PromptBridge sits between the user and the model: it works out what the user is actually trying to do, then rewrites the prompt to match. We built it in a team of four at HackByte 4.0 (IIITDM Jabalpur), where we were among the Top 120 teams in India.

---

What it does

• Detects the intent behind a user's prompt
• Refines the prompt using an LLM (Llama-3-70B on Groq)
• Saves sessions and prompt history so users can go back to earlier versions

Built with: FastAPI, React, PostgreSQL, Groq (Llama-3-70B)

How it's built

Part	What it does
React frontend	Where the user enters a prompt and sees the refined result
FastAPI backend	Receives requests and coordinates the steps below
Intent detection engine	Works out what the user is trying to achieve
Refinement pipeline	Sends the prompt and detected intent to Llama-3-70B on Groq and returns an improved prompt
PostgreSQL	Stores sessions and prompt history

My role

I designed the system architecture and owned the PostgreSQL database design, including connecting the backend to it for session and prompt-history storage. I also helped integrate everything into one working demo. My teammates built the intent-detection engine and the prompt-refinement pipeline.

Run it locally

You'll need Python 3.10+, Node.js 18+, PostgreSQL and a Groq API key.

git clone https://github.com/PrajyotKorde-18/HackByte4.0.git
cd HackByte4.0

Backend

cd backend
pip install -r requirements.txt

Create a .env file:

GROQ_API_KEY=your_key_here
DATABASE_URL=postgresql://user:password@localhost:5432/promptbridge

uvicorn main:app --reload

Frontend

cd frontend
npm install
npm run dev

What I learned

• Designing the data model early made it much easier for four people to build separate parts that still fit together
• Integration at a hackathon is where things break, so agreeing on API shapes up front saves hours
• Working with a team under a deadline means splitting ownership clearly

Team

Built at HackByte 4.0 by a team of four.

• Prajyot Korde: architecture and database design (GitHub)
• Teammates: Naman Chhallani,Yash Dhayal,Durgesh Mundada

Author

Prajyot Korde, IT undergrad at Ramdeobaba University LinkedIn · GitHub