FORTHought

About

FORTHought is a personal, locally-hosted AI research platform specifically configured for Physics and STEM laboratories. It serves as a comprehensive setup, building upon Open WebUI and AMD ROCm, to support a wide array of scientific workflows including literature review, spectroscopy, electron microscopy, X-ray diffraction, and advanced data analysis. This configuration integrates GPU-accelerated LLM inference, a fine-tuned RAG pipeline for scientific documents, and a robust tool orchestration layer via MetaMCP, which aggregates custom-built scientific MCP servers. It focuses on leveraging local hardware for privacy and performance, enabling automated experimental planning, data processing, and document generation without reliance on external cloud services for sensitive research data.

Key Features

• Orchestrates custom scientific MCP servers for automated spectroscopy, XRD, SEM analysis, literature search, and OriginLab workflows.
• Optimized RAG pipeline for scientific documents, enhanced with Docling parsing, Qdrant vector store, and AMD Lemonade reranking.
• Implements agent profiles with skill-gated routing to dynamically expose relevant tools, optimizing token usage.
• Features multi-role local LLMs via AMD Lemonade Server for embeddings, reranking, and vision tasks (VLM).
• 4 GitHub stars
• Integrated AI research platform optimized for Physics/STEM labs, leveraging AMD ROCm for GPU acceleration.

Use Cases

• Automate complex data analysis and fitting workflows for spectroscopy and X-ray diffraction data.
• Accelerate scientific literature review and automate paper retrieval and metadata extraction.
• Aid in planning and modeling photoluminescence experiments and performing electron microscopy image analysis.