Description
A/B test your AI agent prompts in real-time using Supabase session logic and OpenAI. Perfect for optimizing chatbot performance, engagement, or conversion across different prompt styles.
🧠 How It Works:
✅ 💬 Chat Trigger – Messages are received through n8n's chat interface ✅ 🧾 Session Check – Supabase is queried to see if the session ID exists ✅ 🎲 Random Assignment – If it’s a new session, the user is randomly assigned to either a baseline or alternative prompt ✅ 🔄 Session Memory – Session ID and chosen prompt group are stored in Supabase ✅ 🤖 LLM Response – OpenAI generates a response using the assigned prompt variant ✅ 🔁 Persistent Prompting – The same prompt is used throughout the session to maintain context
🔍 It Automates:
✅ A/B prompt assignment using session-based logic ✅ Response generation based on stored prompt group ✅ Session tracking in a live Supabase table ✅ Real-time experimentation without interrupting chat flow
💡 Why Choose This Workflow:
✅ Optimize prompt effectiveness using real user interactions ✅ Run scientific A/B tests without backend changes ✅ Maintain consistent experience per session ✅ Easily extend to test other LLM variables (e.g. temperature, max tokens)
👤 Who Is This For:
✅ Prompt engineers and AI researchers ✅ Product teams running experiments on AI user interfaces ✅ Startups fine-tuning onboarding/chat support bots ✅ Developers building dynamic LLM-powered experiences
🔗 Integrations:
✅ Supabase (for storing session and prompt variant) ✅ OpenAI (for generating AI responses) ✅ n8n Chat or Webhook (for message input and output)
📈 Smarter Prompts Through Real-Time Split Testing Whether you're fine-tuning a support chatbot or experimenting with sales copy, this workflow helps you validate which prompt performs best—scientifically and at scale.
Link : https://lovable.dev/projects/3ce5d6ee-fc85-4465-840d-ea6cbb163b13
