README.md

---
title: Interacting With LLMs
emoji: 💻
colorFrom: green
colorTo: yellow
sdk: gradio
python_version: 3.10.18
sdk_version: 6.5.1
app_file: app.py
pinned: false
license: cc-by-4.0
short_description: Interacting with LLMs
---

Privacy Risk Inference System with Explainability for LLMs: a browser-based chatbot interface that augments a standard conversational agent with real-time privacy risk feedback.
In addition to the standard textual responses of the chatbot, our tool constitutes a layer between the user and the chatbot, proving real-time awareness and explainability of the LLM's inference capabilities.
This layer comprises several real-time detection mechanisms: (1) detection of PIIs (Personally Identifiable Information), (2) linkage against external data pertaining to the user, and (3) profiling mechanism that builds a profile of the user based on the user's conversation and external corpus.
Each mechanism in our threat awareness model generated an explainability feature, which communicates to the user after each conversation turn what information led the LLM to predict a certain profile attribute.
At the end of the conversation with the chatbot, a log with the user's conversation metadata is saved.
The following subsections describe the tool's core components.
PRISE was implemented in Python using the Gradio library, connecting to the underlying LLM through a REST API.