🤖 LogSentinel(JAVA AND PYTHON)

LogSentinel

Dual-Agent AI System for Enterprise IT Troubleshooting

🌟 Overview

LogSentinel is an enterprise-grade troubleshooting assistant that analyzes large error logs mainly focusing on errors of JAVA and PYTHON, retrieves similar historical issues, generates structured step-by-step solutions, and intelligently escalates low-confidence cases.

It combines:

• ScaleDown API for log compression
• Local LLM (Ollama + Mistral) for reasoning
• Embedding-based similarity search
• Confidence-driven escalation workflow

🎯 Key Features

• 🧠 Dual-Agent Architecture (Log Analyst + Solution Engineer)
• 💭Log Compression(via Scaledown API+ Deterministics)
• 🔎 Semantic Similarity Search (StackOverflow, GitHub, Runbooks)
• 📊 Confidence-Based Decision Engine
• 🏠 Local LLM Processing (Sensitive logs stay local)
• 📈 MTTR Tracking & Observability

🚀 Quick Start

Prerequisites

• Python 3.8+
• Ollama installed
• Mistral model pulled
• ScaleDown API Key

Installation

1. Clone the repository

   git clone https://github.com/ishani2025/CHALLENGE-2.git

2. Install dependencies

   pip install -r requirements.txt

3. Install & Start Ollama

   ollama pull mistral
   ollama run  mistral

4. Configure Environment Variables

   Create a .env file and in that have:
   SCALEDOWN_API_KEY=your_key
   SCALEDOWN_BASE_URL=your_url

🏗️ Architecture

LogSentinel
├── src/
│   ├── agents/
│   │   ├── agent1.py
│   │   ├── agent2.py
│   │------llm/
│   ├── embeddings/
│   │   ├── __init__t.py     
│   │   ├── embedder.py       
│   │   └── vector_store.py    
│   ├── utils/
│   │   ├── file_utils.py        
│   │   └── logger.py       
│   └── decision/
│       └── decision_controller.py    
├── data/
│   ├──compressed_logs/              
│   └── raw_logs/            
├── tests/
├──assessts/
└── config/
├── requirements.txt
└── app.py

🤖 Dual-Agent Logic

Agent 1 – Log Analyst

• Calls ScaleDown API
• Reduces noise
• Extracts primary error signals
• Categorizes failure type

Agent 2 – Solution Engineer

• Retrieves similar historical cases
• Generates grounded resolution steps
• Computes confidence score

💬 Confidence Thresholds

   Score     | Action         
   | ------- | -------------- |
   | ≥ 0.85  | Auto Resolve   |
   | 0.65–0.85 | Suggest Review |
   | < 0.65  | Escalate       |

📊 Metrics & Observability

The system tracks:

• Compression Ratio
• Resolution Success Rate
• Escalation Frequency
• Estimated MTTR Reduction

🔧 Running the Application

CLI Mode:

python app.py

Web Interface:

streamlit run app.py

📈 Benefits

• Reduces log size by up to 85%
• Accelerates troubleshooting workflows
• Minimizes unnecessary escalations
• Keeps sensitive data local
• Structured enterprise-aligned pipeline

⚠️ Limitations

• Depends on knowledge base quality
• Confidence scoring requires tuning
• Novel errors may not match historical cases
• Not a full ITSM replacement
• Relies on ScaleDown API availability

🧪 Detailed Presentation: https://docs.google.com/presentation/d/1ukT-LRSfH9DygVu6OdExVY2RyXMrMpR03JLZikN7_p0/edit?usp=sharing