AI-Powered Enterprise PHP Code Generation System

Overview

Royal ERP AI Code Generator indexes legacy PHP ERP codebases, retrieves convention-matched context, parses natural-language requirements, and orchestrates generators for forms, JS handlers, SQL, menus, and installers — then validates and packages the module.

Problem

Hand-writing Royal ERP modules that match house conventions (helpers, vouchers, grids, permissions) is slow and error-prone across a large legacy PHP codebase.

Business challenge

Generate production-ready ERP modules grounded in 1,400+ real company files without modernizing away from legacy ERP patterns.

Solution

Secure ZIP upload → PHP knowledge extraction + vector index → NL requirements → GenerationOrchestrator → ERP validators + optional php -l → downloadable package with feedback capture.

Architecture

Upload → PHP parser / knowledge extract → embed & retrieve similar modules → requirements parse → generators (PHP, functionsJS, SQL, menus, docs) → ERPValidator + static checks → ZIP package.

Implementation

Django 5 + DRF services for analyzer, indexer (VoyageAI or offline embeddings), Anthropic-optional generation with deterministic fallbacks, SSE progress, REST APIs, and Django admin/developer UI.

Challenges

Preserving legacy ERP style while using modern Python orchestration; keeping generation usable offline without API keys.

How we solved them

Deterministic fallbacks for embeddings and generation; Royal-specific validators; optional Anthropic/Voyage when configured.

Security

Path-traversal and symlink protection on ZIP upload, skip rules, authenticated developer access, and validation gates before packaging.

Results

Production code-generation workflow that cuts manual ERP module effort and matches existing conventions at high similarity.

Key modules

• Upload & extractor
• PHP analyzer / knowledge extractor
• Vector indexer & search
• GenerationOrchestrator
• ERPValidator & packaging

Features

• Secure ERP ZIP upload with inventory and module discovery
• Convention-aware PHP parsing and knowledge index
• NL requirements → orchestrated multi-file generation
• PHP forms, functionsJS, SQL schema, menus, installer, docs
• Royal ERP validators + optional PHP lint
• SSE progress, feedback loop, and ZIP download

My responsibilities

• Built end-to-end generation pipeline and orchestrator
• Implemented ERP-specific parsers and validators
• Added indexing, retrieval, and packaging flows

Outcomes

• Grounded on 1,400+ company ERP files
• 70–90% similarity with real codebase conventions
• 60%+ reduction in manual module development effort