Large Language Model

Catch vendor lock-in before it catches you.

FinOps for LLM spend: drop-in OpenAI/Anthropic-compatible proxy with streaming-safe exact cost attribution, waste detection, and LLM-as-judge substitution verdicts. Fails open by design.

RAG 아키텍처 비교 프로젝트 - Naive/Router/Agentic/Cascade RAG 구현 및 성능 비교 (EPL 2024-25) 선수 데이터 기반, LangGraph·FAISS·BGE-M3

Hands-on labs for isolated AI workloads with Azure Container Apps Dynamic Sessions and ACA Sandboxes

RAG system for healthcare billing Q&A — RAGAS-evaluated (1.0 faithfulness, 0% hallucination), deployed live

Fine-tuned 3B model + RAG beats Claude Opus 4.8 raw at entity resolution (99% vs 78.5%) on unseen companies — Covent LLM Challenge submission.

Transcribe & repurpose video content (URL/file) into summaries, chapters, captions and more — powered by Whisper & Ollama

MVP RAG scalable et souverain — chatbot d'événements culturels : Mistral, pgvector, PostGIS, Redis, agents smolagents, monitoring Langfuse, cible Scaleway (cloud souverain UE).

Ask questions about your PDFs and get cited answers — or an honest "not found." Built from scratch in Python.

终端 AI Agent CLI — ReAct/Plan-and-Execute/Multi-Agent/MCP/RAG

Smart IT support ticket router using FastAPI, RAG, pgvector semantic search, and Redis background queues.

An AI-powered PDF Question Answering application using LangChain RAG architecture, FAISS vector database, Hugging Face embeddings, OpenAI GPT-4o Mini, and Streamlit.

Portfolio: multi-tier LLM gateway, cost ledger, tool-calling agent, eval/regression harness

AI Contract Intelligence Platform — RAG-powered natural-language Q&A, automated clause-risk flagging, and obligation-deadline tracking over enterprise contracts. FastAPI + React + PostgreSQL + ChromaDB + Ollama (Mistral 7B), fully local and free via Docker Compose.

Multi-provider LLM usage, cost & rate-limit meter with budget guards — Anthropic/OpenAI/OpenRouter/local, TUI + MCP + CI exit codes. Zero-dependency.

Boilerplate for building local LLM apps with Ollama

An autonomous, Multi-Agent AI system for Anti-Money Laundering (AML) that fuses deterministic structuring rules with unsupervised ML (PyOD) to dynamically detect financial crime and auto-generate explainable SARs.

PRISM: Pre/post-inference Runtime Inference Safety Monitor

Local-LLM RAG chat assistant for internal documentation — hybrid retrieval, citation-backed answers, zero cloud LLM calls. FastAPI + React, Docker Compose (Mac/GPU/AWS) + k8s stubs.

Distributed GPT-2 Large pretraining with PyTorch DDP across multi-node H100s — throughput, NCCL communication overhead, and scaling efficiency analysis on Nebius AI Cloud.