Skip to content

LucyDetect is an LLM Drift Analyzer designed to help researchers track AI model responses over time and detect inconsistencies, including hallucinations. It stores past responses, calculates drift scores, and alerts users to significant changes in model behavior.

License

Notifications You must be signed in to change notification settings

rainbowgore/LucyDetect

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🔮 LucyDetect - LLM Drift Analyzer

📌 Overview

LucyDetect helps developers and AI researchers analyze LLM response drift over time. It tracks and compares responses from GPT models, measuring consistency and detecting unexpected changes.

Why Does This Matter? (Industry Gap)

🔍 The Problem: No Standard for Monitoring LLM Consistency

Right now, AI teams, researchers, and businesses have no standard way to:

  1. Detect when an LLM response changes over time.
  2. Measure how different two LLM-generated answers are.
  3. Track drift after fine-tuning, retraining, or updates.
  4. Compare two LLM models (e.g., GPT-4o vs. GPT-3.5) for reliability.
  5. Prevent AI hallucinations and ensure factual consistency.

🛠 How LucyDetect Fills the Gap

LucyDetect quantifies response drift by: ✔ Logging all GPT responses in a database.
Using FAISS for similarity search to compare past answers.
Computing drift scores based on semantic similarity.
Visualizing trends over time with an interactive UI.


🚀 Features

Detect Response Drift → Compare past and current LLM outputs.
Log & Analyze Trends → Track drift scores over time.
Store & Retrieve Data → Securely save responses for future comparison.
Custom API Key Support → Users enter their own OpenAI API key.
Web UI (Streamlit) → Easy-to-use interface for drift detection.


Use Cases

👩‍💻 For AI Developers

  • Monitor model consistency across deployments.
  • Detect drift when fine-tuning or retraining LLMs.
  • Benchmark different OpenAI models before production use.

🏢 For Businesses

  • Ensure AI-generated responses stay consistent for users.
  • Prevent sudden answer changes in customer-facing applications.

🔬 For Researchers

  • Study long-term model behavior under different conditions.
  • Investigate semantic drift in language models over time.

How to Use LucyDetect

1️⃣ Use the Public Web UI (No Installation Required)

Anyone can use LucyDetect online without setup!
🔗 **LucyDetect Web UI


2️⃣ Run Locally (For Developers)

Clone the repo and install dependencies:

git clone https://github.com/your-repo/lucydetect.git
cd lucydetect
pip install -r requirements.txt

About

LucyDetect is an LLM Drift Analyzer designed to help researchers track AI model responses over time and detect inconsistencies, including hallucinations. It stores past responses, calculates drift scores, and alerts users to significant changes in model behavior.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages