dcostenco/prism-coder-14b
TEXT GENERATIONConcurrency Cost:1Model Size:14.8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold
dcostenco/prism-coder-14b is a fine-tuned Qwen3-14B model developed by dcostenco, specialized for tool routing and healthcare-specific TypeScript code generation. The model excels at routing to 17 Prism Memory tools with 99.7% accuracy on a 300-case evaluation and achieves 100% on a 22-check Synalux healthcare TypeScript coding evaluation. It is designed for applications requiring precise tool orchestration and domain-specific code generation within healthcare platforms.
Loading preview...
dcostenco/prism-coder-14b: Specialized Tool Router and Healthcare Coder
dcostenco/prism-coder-14b is a Qwen3-14B model fine-tuned by dcostenco, primarily for the Prism AAC / Synalux healthcare platform. It features two main specializations: a highly accurate tool router and a healthcare-specific TypeScript code generator.
Key Capabilities
- Prism Memory Tool Routing: The production S14 version demonstrates exceptional performance, achieving 99.7% strict accuracy on a 300-case evaluation involving 17 distinct Prism Memory tools. This includes precise routing for functions like
session_save_ledger,session_search_memory, andknowledge_search, with only one ambiguous edge case identified. - Healthcare TypeScript Code Generation: A legacy version (v42) achieved 100% accuracy across 22 checks for generating production-ready Next.js API routes for X12 835 ERA reconciliation against 837P claims, incorporating critical features like HIPAA compliance, audit wrappers, and cross-tenant guards.
- Robust Training: The S14 router was developed through a surgical LoRA chain (S1→S14) on Qwen3-14B, using MLX LoRA with specific parameters and a balanced corpus of natural-phrasing and tool-use SFT.
- System Prompt Integration: Utilizes a detailed system prompt defining its identity as Synalux, a memory-augmented coding and clinical reasoning assistant, with access to 17 Prism Memory tools and 13 multimodal tool modules.
Good for
- Automated Tool Orchestration: Ideal for applications requiring precise and reliable routing to a predefined set of tools, particularly in complex, multi-step workflows.
- Healthcare-Specific Development: Highly suitable for generating accurate and compliant TypeScript code within the healthcare domain, especially for tasks involving claims processing and HIPAA-sensitive data.
- Integration into Larger AI Systems: Can serve as a critical component in cascaded AI architectures, handling production routing for complex reasoning tasks.