Params.AI

A research platform comparing chat vs form interfaces for prompt reuse in LLM applications

What is This?

Params.AI tests whether structured forms are better than chat for reusing AI prompts. Think of it as: "What if instead of copy-pasting prompts into ChatGPT, you could turn them into reusable apps with simple input forms?"

The Research Question

When you create a useful AI prompt, how do you reuse it later? We're comparing two approaches:

• Chat Interface (Baseline) - Copy-paste and modify prompts manually, like ChatGPT
• Form Interface (Experimental) - Convert prompts to {{variable}} templates that auto-generate forms

Example:

Original: "Write a professional email to my boss about the Q4 report"

Template: "Write a {{tone}} email to {{recipient}} about {{subject}}"
         → Auto-generates form with 3 input fields

Quick Start (20 minutes)

Prerequisites

• Node.js 18+
• pnpm 8.15.4+ (npm install -g pnpm)
• Firebase account (free tier)
• OpenAI API key or similar LLM provider

Installation

# 1. Clone and install
git clone <repository-url>
cd params-ai
pnpm install

# 2. Set up environment files (see SETUP.md for details)
cp packages/backend/.env.example packages/backend/.env
cp packages/frontend/.env.example packages/frontend/.env.development
# Edit .env files with your credentials (Firebase, database, LLM API keys)
# Copy Firebase service account JSON to packages/backend/src/config/firebase-key.json

# 3. Configure database (PostgreSQL or Neon)
cd packages/backend
pnpm prisma migrate dev
pnpm prisma db seed

# 4. Start development servers
cd ../..
pnpm dev

First time setup? See SETUP.md for detailed step-by-step instructions.

Access Points

• Frontend: http://localhost:5173
• Backend API: http://localhost:8080 (serves both REST at /api/v1 and tRPC at /trpc)
• Database Studio: http://localhost:5555 (run pnpm prisma studio in backend)

Project Status

✅ What Works

• Form-based interface - Fully functional prompt lab with variable extraction
• Multi-LLM support - OpenAI, Google Gemini, Cloudflare Workers AI
• Experiment framework - Database schema and admin UI for setting up studies
• Template system - Generalize prompts into reusable templates

❌ What's Missing (Priority for Contributors)

• Chat interface - Baseline condition for comparison study
• Reuse workflows - Both chat-based (Reuse 1) and form-based (Reuse 2) flows
• Metrics tracking - Timing, clicks, task completion rates
• Data export - CSV/JSON for analysis

How It Works

For Researchers

1. Create experiments comparing chat vs form interfaces
2. Recruit participants and track their workflows
3. Export data for statistical analysis

For Creators (App Building)

1. Write prompts with {{variables}} in the Prompt Lab
2. Platform auto-generates interactive forms
3. Test and iterate in real-time
4. Publish apps for others to use

For End Users

1. Browse published apps
2. Fill simple forms (no prompt engineering needed)
3. Get instant AI-generated results

Architecture

graph TB
    subgraph Frontend ["Frontend (React + Vite)"]
        A[Prompt Lab<br/>3-column editor] 
        B[Published Apps<br/>2-column UI]
        C[Experiment Tasks<br/>Research workflow]
    end
    
    subgraph Backend ["Backend (Node.js + tRPC)"]
        D[tRPC Router<br/>Type-safe API]
        E[Prisma ORM]
    end
    
    subgraph External ["External Services"]
        F[(PostgreSQL<br/>Neon)]
        G[LLM Providers<br/>OpenAI, Google, etc]
        H[Firebase Auth]
    end
    
    A --> D
    B --> D
    C --> D
    D --> E
    E --> F
    D --> G
    Frontend --> H

Loading

Detailed architecture? A dedicated architecture doc is planned; meanwhile the new User Guide covers current UI flows and data touchpoints.

Tech Stack

Frontend: React 18, TypeScript, Vite, TailwindCSS, CodeMirror, Mustache.js
Backend: Node.js, Express, tRPC, Prisma
Database: PostgreSQL (Neon serverless recommended)
Auth: Firebase Authentication
LLM: OpenAI, Google Gemini, Cloudflare Workers AI
Monorepo: pnpm workspaces + Turborepo

Project Structure

params-ai/
├── packages/
│   ├── frontend/        # React app with Prompt Lab and experiments
│   ├── backend/         # tRPC API + Prisma database
│   └── llm/             # LLM provider abstraction layer
├── docs/                # Documentation
│   ├── SETUP.md         # Step-by-step setup guide
│   └── user.md          # User flows mapped to the current implementation
└── README.md            # This file

Common Commands

# Start both frontend and backend
pnpm dev

# Run only frontend or backend
pnpm dev:frontend
pnpm dev:backend

# Database operations
cd packages/backend
pnpm prisma studio          # Visual database browser
pnpm prisma migrate dev     # Create/apply migrations
pnpm prisma db seed         # Add sample data

# Build for production
pnpm build

Documentation

• Setup Guide - Detailed installation instructions with troubleshooting
• User Guide - Researcher/participant/creator/end-user flows mapped to the current code

Architecture, Research, Contributing, and API docs are planned; add an issue if you need them sooner.

More help: See SETUP.md for comprehensive troubleshooting.

Research Context

This platform supports comparative studies of prompt reuse interfaces. The core hypothesis: structured form-based interfaces may improve prompt reusability compared to conversational chat interfaces.

Key metrics being studied:

• Time to create and generalize prompts
• Success rates of prompt reuse
• User preference and cognitive load
• Task completion rates

Research methodology: Document coming soon—capture interim notes in docs/user.md or issues.

---

Status: Active research project | Contributors welcome | For academic and non-commercial use