nomeda-lab/Fattah-Orch-Small

Fattah-Orch-Small: Arabic-First Coding Orchestrator

Fattah-Orch-Small, developed by Nomeda Lab, is a 1.7 billion parameter model within the Fattah-Orch family, designed to act as an intelligent orchestrator for AI coding pipelines. It accepts software requests in Egyptian Arabic, Modern Standard Arabic, or English and translates them into a structured, typed, and dependency-ordered JSON task graph. This process aims to provide clear, precise instructions to subsequent coder models (like GPT-4o or Claude), reducing ambiguity and improving output quality.

Key Capabilities

• Multilingual Input: Understands software requests in Egyptian Arabic, Modern Standard Arabic, and English.
• Task Orchestration: Breaks down high-level requests into a sequence of granular, executable subtasks.
• Structured Output: Generates a JSON task graph with defined request_summary, subtasks (including id, title, description, type, and depends_on).
• Supported Task Types: Can define tasks for various programming languages and domains, including python, typescript, sql, go, kotlin, swift, and bash.
• Efficiency: By pre-processing requests, it helps reduce token usage and back-and-forth interactions with larger, more expensive coder models.

Use Cases and Limitations

Fattah-Orch-Small is ideal for developers looking to streamline their AI-assisted code generation workflows, especially when dealing with Arabic-language requirements. It acts as a crucial intermediary, ensuring that complex software ideas are systematically decomposed before code generation. While it excels at planning, it does not write code itself; it must be paired with a separate coder model. Performance on Egyptian Arabic is noted as best, with other Arabic dialects and English also supported. For very large or complex systems, the larger Fattah-Orch-M or Fattah-Orch-L models are recommended for richer task descriptions.