Agentic LLM

A curated Agentic LLM suite regrouping modular, production-oriented building blocks for real-world Large Language Model systems.

This repository acts as a parent container for multiple independent sub-projects, each focused on a specific capability of agentic LLM systems.

Repository Structure

• autonomous-ai-platform
  Complete agentic LLM platform supporting local models, GPU inference via vLLM, RAG pipelines, Text-to-SQL, evaluation, and Airflow orchestration.

• llm-proxy-gateway
  LLM routing and proxy layer for model selection, policies, observability, and cost control.

• rag-drive-gcp
  Retrieval-Augmented Generation pipeline built on Google Drive, OCR, GCS, Vertex AI, and Streamlit.

• local-finetuning
  Local fine-tuning workflows for domain adaptation, dataset preparation, and evaluation.

• local-quantization
  Quantization pipelines for efficient local inference (CPU / GPU).

Each sub-project is:

• self-contained
• independently testable
• versioned in its own repository
• integrated here using git subtree

Philosophy

This suite is designed with:

• clean architecture
• explicit configuration
• production-grade error handling
• minimal coupling between components

The goal is to demonstrate end-to-end agentic LLM system design, not isolated scripts.

Contact

GitHub: https://github.com/GeorgesNass
LinkedIn: Georges Nassopoulos