🟢 Slide 3 — Problem Statement
Nowadays most companies use cloud platforms to store sensitive data.
But current cloud security systems are mainly rule-based, which means they only detect
known attacks.
The main problem is —
They cannot detect new or unknown attacks, especially zero-day attacks.
Cyberattacks are increasing.
Manual monitoring is slow and costly.
False alerts are high in current systems.
So we need an intelligent, automated, real-time security system.
Our project focuses on:
Real-time log monitoring
AI-based anomaly detection
Automatic threat alerts
Scalable cloud security solution
🟢 Slide 4 — Literature Survey
We studied multiple research areas like:
AI-based intrusion detection
Machine learning anomaly detection
Deep learning cybersecurity models
Transformer and GPT-based analysis
Existing Techniques:
Rule-based → only known attacks
Signature-based → fast but limited
ML models → need training data
Deep learning → accurate but heavy computation
Research Gaps:
No real-time Generative AI detection
High false positives
Cannot detect unknown attacks
Poor context understanding
🟢 Slide 5 — System Architecture
Our system follows this pipeline:
Cloud Logs → Preprocessing → GPT Analysis → Detection → Alert → Dashboard
Modules:
Log Collection
Data Preprocessing
GPT Analysis
Detection Engine
Alert System
Dashboard Visualization
This ensures continuous real-time monitoring.
🟢 Slide 6 — Proposed Methodology
Our methodology works step-by-step:
First, we collect cloud logs.
Then we preprocess and clean the data.
Then GPT analyses log behaviour patterns.
Then anomaly detection classifies normal vs suspicious.
Then alerts are generated and stored.
Models Used:
GPT / Transformer models
Threshold-based anomaly detection
(Optional) Ensemble methods
🟢 Slide 7 — Implementation Details
So far we have implemented:
Data collection pipeline
Preprocessing module
Initial GPT integration
Prototype detection module
MongoDB storage setup
Basic dashboard UI
Tools Used:
Python
FastAPI / Flask
TensorFlow
GPT API
MongoDB
Kafka / RabbitMQ
AWS Cloud
🟢 Slide 8 — Dataset and Design
We are using:
Public cybersecurity datasets
Cloud log datasets like AWS CloudTrail
Preprocessing Includes:
Missing value removal
Normalization
Encoding categorical values
Design Focus:
Real-time processing
High scalability
Secure data handling
🟢 Slide 9 — Results and Observations
Current system results:
Log collection working successfully
Preprocessing pipeline stable
AI detecting abnormal patterns
Alerts generating correctly
Performance:
Accuracy around 85–90% currently
Near real-time detection achieved
Observation:
AI detects complex attacks better than rule-based systems.
🟢 Slide 10 — Challenges Faced
Technical Challenges:
High GPU requirement → Used cloud GPU
Real-time data processing → Used Kafka streaming
Log format differences → Standardized JSON
Data Challenges:
No real attack datasets → Used simulated attack logs
Imbalanced data → Used balancing techniques
Noisy logs → Applied cleaning pipeline
🟢 Slide 11 — Review 2 Status
At this stage:
Literature survey is completed
System design is completed
Partial implementation is completed
System is ready for full development phase.
🟢 Slide 12 — Future Work
Next steps include:
Completing all modules
Full dashboard development
Model accuracy improvement
Real-time production testing
Performance optimization
Final documentation
🟢 Slide 13 — Conclusion
In Review 2, we completed research, system design, and initial implementation.
We successfully built preprocessing and initial AI detection modules.
Our project shows strong potential for real-time cloud security using Generative AI.
We are ready to move to full implementation and testing phase for final review.
Thank you.