F

finetuning

AI Development Platform. Create Datasets - Train - Monitor - Share

LogeIon

LogeIon is an open AI development and orchestration platform designed to build, fine-tune, evaluate and monitor domain-specific intelligent systems.

Originally developed for advanced Legal AI research, LogeIon is evolving into a modular platform for creating custom AI systems across any field — law, medicine, engineering, finance, science, education and beyond.

The platform combines:

AI fine-tuning pipelines Retrieval-Augmented Generation (RAG) Multi-model orchestration Dataset generation workflows Real-time monitoring dashboards Evaluation and benchmarking systems GPU and inference management Local-first and hybrid cloud AI infrastructure ⸻

Core Vision

LogeIon aims to provide developers, researchers and organizations with full control over the AI lifecycle:

collect knowledge generate datasets train models evaluate reasoning monitor performance deploy specialized AI systems without relying on opaque closed ecosystems.

⸻

Technology Stack

FastAPI orchestration backend Qdrant vector database Ollama / vLLM / llama.cpp inference Transformers / LoRA / Unsloth fine-tuning LangGraph AI workflows Real-time monitoring & analytics Modular local-first architecture ⸻

Current Focus

Italian Legal AI Retrieval-enhanced reasoning Citation-aware generation Hallucination reduction Autonomous dataset pipelines AI evaluation frameworks Multi-stage training orchestration ⸻

Long-Term Direction

LogeIon is evolving beyond Legal AI into a general-purpose AI development platform capable of powering highly specialized reasoning systems for any domain.

The goal is to create a transparent, modular and extensible infrastructure for the next generation of open intelligent systems.

Fine-tune SeaLLM-7B / Llama-3-Thai ด้วย QLoRA

Check out the project preview: https://imdb-movie-classifier-76ac6b.gitlab.io/

moostral is a fine-tuning on Mistral 7B with the goal of translating multi-objective optimization problems from a textual representation to a jMetal implementation automatically.