Home Pricing Archive About Privacy

LLM-Augmented Tamper-Evident Secure Logging Framework

📑 10 slides 👁 22 views 📅 1/30/2026
0.0 (0 ratings)

Title Slide

Project title: LLM-Augmented Tamper-Evident Secure Logging Framework

Title Slide
2

Abstract

  • Addresses log tampering and PII leakage using local LLM and Merkle chaining
  • Novel integration of Phi-3 LLM for real-time anomaly detection
  • Features PII masking, tamper-evident logs, and ntfy notifications
  • Benefits: Enhanced security, privacy, and real-time monitoring
Abstract
3

Introduction

  • Secure logging is critical in AI/enterprise systems to prevent breaches
  • Problem: Log tampering, PII exposure, lack of real-time detection
  • Motivation: Incidents like SolarWinds highlight need for secure logs
  • Overview: System ensures log integrity and privacy with AI analysis
Introduction
4

Scope

  • Included: Local deployment, real-time anomaly detection, PII masking
  • Excluded: Cloud integration, multi-user auth, production scaling
  • Focus on lightweight, local solution with AI-powered intelligence
  • Notifications via ntfy for immediate alerting
Scope
5

Objectives

  • Primary: Build tamper-evident logging with privacy and AI
  • Implement Merkle-like chained hashing for integrity checks
  • Integrate local Phi-3 LLM via Ollama for anomaly detection
  • Provide real-time alerts and user-friendly React dashboard
Objectives
6

System Architecture

  • Flow: User → React UI → FastAPI → PII Masking → LLM → Hashing
  • Storage in SQLite with integrity verification via Merkle chaining
  • Real-time alerts dispatched via ntfy.sh
  • Trust boundaries clearly marked in data flow
System Architecture
7

Implementation

  • Technologies: React, FastAPI, SQLite, Ollama, Docker, ntfy
  • Steps: Setup Docker, build API, integrate LLM, add UI, notifications
  • Key challenge: Ensuring low-latency LLM analysis for real-time logs
  • Result: Containerized, scalable system with modular components
Implementation
8

Modules

  • Frontend: Dashboard, Logs, Alerts, ChainBanner, DemoButtons
  • Backend: Ingestion API, Test Endpoints, ntfy Dispatch
  • AI: Ollama + Phi-3 container for local processing
  • Storage: SQLite logs table with hashed entries
Modules
9

Related Work

  • Compared to ELK, Splunk, Acra - lacks local LLM and lightweight design
  • Block-DEF uses heavy blockchain; our Merkle chaining is efficient
  • SecureLLM requires cloud; our solution is fully local
  • Novelty: Combines chaining, LLM, notifications in one system
Related Work
10

Conclusion

  • Achieved: Tamper-evident logs with PII protection and AI analysis
  • Key benefits: Privacy, real-time alerts, local processing
  • Future: Multi-node support, advanced LLM fine-tuning, RBAC
  • Demonstrates viable alternative to cloud-dependent logging tools
Conclusion
1 / 10