Report Abuse

Basic Information

Tianji is an open source project for building and experimenting with large language model applications focused on social etiquette and interpersonal scenarios. It bundles end-to-end resources for developers and researchers including prompt engineering collections, retrieval-augmented generation (RAG) knowledge bases, agent implementations, model fine-tuning recipes, and data manufacturing and cleaning tools. The repository provides runnable demos and frontends for prompt, RAG and agent modes, example scripts to call various online and local LLMs, and guidance on environment and API key configuration. It also publishes curated datasets and fine-tuned models for specific social scenarios and contains documentation and tutorials aimed at teaching full-stack LLM application development for vertical domains.

Links

Categorization

App Details

Features
The project supports multiple model backends and invocation modes, with examples for ChatGPT, ZhipuAI, ERNIE, InternLM, Qwen and both online and local calls. It provides RAG implementations using LangChain and LlamaIndex and tooling to build and query knowledge bases. Agent applications are implemented and demoed, including MetaGPT-based agents and a Streamlit frontend. Fine-tuning workflows include Lora and full-parameter training examples using frameworks like Transformers and Xtuner. Data manufacturing and cleaning tools and curated social-etiquette corpora are included and published to Hugging Face. The repo contains runnable scripts, demo frontends, environment variable guidance, and CI-friendly contribution flows for prompt updates.
Use Cases
Tianji helps developers learn and reproduce a complete pipeline for creating domain-specific LLM apps by providing concrete code, datasets and tutorials. It lowers the barrier to experiment with prompt-only systems, RAG-based knowledge retrieval, agent orchestration and model fine-tuning by packaging example projects, demo frontends and configuration patterns. Teams can reuse the provided social etiquette corpora, data cleaning scripts and fine-tuning configs to adapt models for their own verticals. The repository also offers contribution guidance for prompt additions and a set of runnable demos to validate model behavior, making it practical for both learning and bootstrapping production prototypes in conversational and agent-driven applications.

Please fill the required fields*