AI-Assisted Liferay Code Generator with RAG

AI-Powered Dual Code Generation

A Streamlit-based assistant was developed to facilitate dual-mode generation of Java and Groovy code tailored to the Liferay 7.4 DXP environment:

• Developers interact via a conversational UI
• Code generation triggered for both Java modules and Groovy scripts
• Outputs aligned with Liferay coding standards and configurations

Retrieval-Augmented Generation (RAG) Pipeline

To improve contextual accuracy:

• The Liferay 7.4 source codebase was indexed using Sentence Transformer embeddings
• A Chroma DB vector store supported fast semantic retrieval of relevant code snippets
• Retrieved context-enriched prompt inputs sent to the LLM
  This ensured that the generated outputs reflected real code patterns and standards. 

Smart Prompt Engineering

Advanced prompts were created to ensure code outputs:

• Follow strict Liferay conventions
• Adapt dynamically to Java or Groovy requirements
• Leverage modular prompt files for flexible AI instruction
  This approach minimized unpredictable outputs and improved compliance with enterprise requirements. 

Developer-Friendly Interface

The assistant included:

• A branded Streamlit UI with toggles between Java and Groovy modes
• Support for persistent session history

Custom styling and dark mode for enhanced usability
This helped developers explore, revisit, and refine generated content during sessions.

• Access Control: OAuth2 and role-based access to the assistant
• Encrypted Keys & Tokens: Secure storage and use of API credentials
• Traceability: Persistent session history and audit logs
• Governance: Prompt templates enforcing coding standards and compliance

These measures ensured controlled, auditable AI code generation aligned with enterprise and developer governance.