Skip to main content

Generic LLMs Fail on BBj

Intelligent Code Assistance Across Four Generations of Business BASIC

BBj spans four decades of Business BASIC evolution, but virtually none of that code exists in public training data. Copilot hallucinates. ChatGPT guesses. A custom AI strategy is the only path forward.

Read the Strategy

Why This Strategy Exists

BBj is the modern face of a language lineage that began in the 1980s with Business BASIC. Over four generations -- BBx, PRO/5, Visual PRO/5, and BBj -- the platform evolved from character-mode terminal applications to browser-based DWC apps running on the JVM. Thousands of production systems still rely on this stack, many carrying code from multiple generations in the same codebase.

Large language models trained on public data have effectively zero coverage of BBj. When developers ask Copilot or ChatGPT for help, they get plausible-looking code that uses wrong syntax, invents nonexistent functions, and confuses generation-specific APIs. The hallucination rate is not a minor inconvenience -- it makes generic AI tools actively harmful for BBj development.

This strategy proposes three integrated initiatives, built on a shared AI infrastructure, to deliver real code intelligence for BBj developers.

Fine-Tuned BBj Model

A LoRA-trained language model hosted via Ollama that understands all four generations of BBj -- from legacy Business BASIC to modern DWC applications.

IDE Integration

A VSCode extension powered by Langium that delivers intelligent completions, diagnostics, and generation-aware suggestions directly in the editor.

Documentation Chat

A RAG-powered conversational interface embedded in this documentation site, providing generation-aware answers grounded in real BBj knowledge.

Find What Matters to You

For Developers

How the fine-tuned model, IDE extension, and RAG pipeline work -- technical depth for implementers.

For Leadership

Strategic vision, architecture decisions, and implementation roadmap -- the business case for BBj AI.

For Customers & Partners

What this means for your BBj applications -- the developer experience improvements ahead.

All Chapters

1

The BBj Challenge

Why generic LLMs fail on four generations of Business BASIC.

2

Strategic Architecture

A unified infrastructure powering all AI capabilities.

3

Fine-Tuning the Model

Training a BBj-aware language model with LoRA and Ollama.

4

IDE Integration

Bringing BBj intelligence into the developer's editor.

5

Documentation Chat

Generation-aware answers embedded in the docs site.

6

RAG Database Design

Multi-generation retrieval for precise, context-aware results.

7

Implementation Roadmap

Progress summary and forward plan for the BBj AI strategy.