System Architecture.
Examine service boundaries, API contracts, data flows, and infrastructure components. Click to zoom and pan.
Demo Walkthrough.
Watch an end-to-end execution of the system in action.
Runtime Screenshots.
Interface states, developer logs, and monitoring views. Click any image to view in lightbox.
Technical Specifications.
Verified parameters and operational benchmarks for this release.
Tech Stack.
Features Checklist.
Core capabilities, verified deliverables, and technical assets included with this release.
- Prompt injection detection
- Jailbreak detection
- System prompt protection
- PII detection and redaction
- Aadhaar detection
- PAN detection
- UPI detection
- Input and output scanning
- Custom guardrail rules
- Multi-stage threat detection
- REST API interface
- LLM-agnostic architecture
- Configurable security policies
- Threat/risk classification
- API integration examples
- Docker-based deployment
- Logging and debugging support
- Production-oriented project architecture
System Requirements.
- OS
- Linux / macOS / Windows with WSL2
- CPU
- 4 cores minimum
- GPU
- Not required for basic deployment
- RAM
- 8 GB minimum, 16 GB recommended
- Docker
- Docker Engine / Docker Desktop recommended
- Python
- 3.10+
- Storage
- 5 GB+
- Internet
- Required for external LLM API integrations
About this Project.
Build an AI Security Layer for LLM Applications
Rakshak is a production-oriented runtime guardrail system designed to sit between an application and its Large Language Model (LLM).
It screens incoming prompts and outgoing model responses for common AI-specific security risks such as prompt injection, jailbreak attempts, system-prompt extraction, and sensitive-data leakage.
This Artifactly project gives you the opportunity to study and run the architecture behind a real AI security system built by Kalpit Labs.
What you'll build
Rakshak provides a security layer between your application and its LLM:
User → Rakshak → LLM → Rakshak → User
Incoming requests can be inspected before they reach the model, while model responses can be inspected before they are returned to the user.
Key capabilities
- Prompt injection detection
- Jailbreak detection
- System prompt protection
- PII detection and redaction
- Aadhaar, PAN and UPI detection
- Input and output security scanning
- Configurable guardrail policies
- Custom security rules
- Risk/threat classification
- REST API integration
- LLM-agnostic architecture
- Docker-based deployment
What you will learn
This project is designed for developers and students who want to understand how AI security works beyond basic prompt engineering.
You can study:
- How LLM applications can be attacked
- How prompt injection detection works
- How input and output guardrails can be implemented
- How sensitive information can be detected and protected
- How middleware can be placed between an application and an LLM
- How AI security controls can be structured as an API
- How to design a security-oriented production system
Who is this for?
- AI/ML developers
- Cybersecurity students and professionals
- LLM application developers
- AI security researchers
- Final-year project students
- Developers building AI agents and assistants
- Anyone interested in production AI security
Why this project is different
This isn't a basic chatbot or a tutorial implementation.
Rakshak is based on a real AI security system developed by Kalpit Labs and designed around security problems encountered in real LLM applications.
The project is intended to help you understand how a production-oriented AI security layer is structured, deployed, tested and integrated.
Important
This Artifactly project is provided for learning, experimentation, development and permitted commercial use according to the included licence.
It does not automatically include access to Kalpit Labs' hosted Rakshak service, managed infrastructure, or commercial API plans.
Built by Kalpit Labs
Rakshak is an original system developed by Kalpit Labs as part of its AI security infrastructure work.