Create new README.md

README.md

@@ -1,147 +1,74 @@
-# 🧠 AI Novelist RAG
+# AI Novelist RAG · Long-Form Story Generation with Memory
 
-A dedicated AI system for long-form novel generation, designed to tackle common issues like **incoherent logic**, **self-contradiction**, and **theme drift** in LLM-generated narratives.
+This is a dedicated AI system for generating long-form novels with coherent logic, consistent world-building, and thematic integrity, solving common LLMs issues like incoherence, self-contradiction, and theme drift.
 
-This project integrates **Langchain + RAG + FastAPI + Local Quantized Models (e.g., Qwen)** to enable the AI to **extract key narrative elements** from its own outputs and store them in a memory-enhancing vector database (FAISS), effectively boosting logic retention and narrative consistency.
+This project combines Langchain, Retrieval-Augmented Generation (RAG), FAISS, FastAPI, and Gradio, along with OpenAI or local quantized models (e.g., LLaMA). It allows the AI to extract key information and context from its own outputs and store them in a vector database to “remember” what it’s written—resulting in more logical and immersive storytelling. 
 
----
+This system is designed as both a technical practice and a creative tool. 
 
-## 🧩 Core Idea
+## Key Features
 
-> **Let the AI "remember" what it writes, to keep stories logically consistent.**
+- High-quality novel generation (via OpenAI APIs or local LLMs)
+- Automatic information extraction (currently via BERT-based summary model)
+- Long-context memory (via FAISS vector store)
+- Context-aware coherent prompt (via RAG-like context enhancement)
+- Web-based UI (via Gradio)
+- Experimental consistency scoring (planned to use long-context metrics)
 
-- ✅ High-quality novel generation via Qwen
-- ✅ Automatic extraction of key narrative settings (characters, themes, backgrounds)
-- ✅ Vector memory with FAISS for context recall
-- ✅ Retrieval-augmented writing with self-referencing history
-- ✅ Optional consistency scoring using CoT-style benchmarks
-
----
-
-## 📂 Project Structure
+## Project Structure
 
 ```
 ai-novelist-rag/
 ├── app/
-│   ├── main.py
-│   ├── config.py
-│   ├── apis/
-│   │   ├── generator.py
-│   │   ├── extractor.py
-│   │   ├── memory.py
-│   │   └── benchmark.py
-│   ├── models/
-│   │   ├── model.py
-│   │   └── faiss_index.py
-│   ├── chains/
-│   │   ├── rag_chain.py
-│   │   └── memory_insert_chain.py
-│   └── utils/
-│       ├── logger.py
-│       ├── text_processing.py
-│       └── utils.py
-├── data/
-│   ├── memory_store.faiss
-│   └── samples/
-│       ├── raws/
-│       └── processed/
-├── docker/
-│   ├── Dockerfile
-│   └── docker-compose.yml
+│   ├── main.py                # FastAPI + Gradio mount point
+│   ├── apis/                  # API endpoints for generation, editing, benchmarks
+│   ├── front_end/             # Gradio-based UI
+│   ├── managers/              # Managers and chains for chapter, summary, and vector memory
+│   ├── models/                # Model loading and wrapper logic
+│   ├── tests/                 # Jupyter notebooks for quick testing
+│   └── utils/                 # General utilities 
+│
+├── data/                      # (Not showed here) Local store for texts, summaries, vector DB 
+├── docker/                    # Dockerfile and config for deployment
 ├── scripts/
-│   └── launch_local.sh
-├── tests/
+│   ├── setup.sh               # Install dependencies
+│   └── launch_local.sh        # Launch the app locally
+│
 ├── requirements.txt
 ├── README.md
 └── .gitignore
 ```
 
----
+## Getting Started
 
-## 🚀 Quick Start
+1. Install dependencies:
 
 ```bash
-bash scripts/launch_local.sh
+bash scripts/setup.sh
 ```
 
-Default: Local quantized Qwen model (configurable via `config.py`).
-
----
-
-## 📌 Tech Stack
-
-- 🤖 Qwen quantized (INT4 / bfloat16)
-- 🧱 Langchain pipeline management
-- 🔍 FAISS for vector search
-- 🔁 Retrieval-Augmented Generation (RAG)
-- 🛠️ FastAPI for backend serving
-- 🐳 Docker/docker-compose (optional deployment)
-
----
-
-## 📌 TODO
-
-- [ ] Interactive multi-turn topic-driven generation
-- [ ] CoT-style logic scoring feedback loop
-- [ ] Azure / Hugging Face Spaces deployment
-- [ ] Unit test integration
-
----
-
-## 🧠 Suitable For
-
-- AI + Literature applications
-- Multimodal + long-context logic reasoning
-- Langchain + RAG stack practice
-- Technical portfolio for interviews
-
----
-
-# 🧠 AI Novelist RAG（中文）
-
-一个专为“小说生成”设计的 AI 系统，致力于解决大语言模型在长文本生成中常见的 **逻辑性弱**、**前后矛盾**、**主题偏移** 等问题。
-
-本项目结合 **Langchain + RAG + FastAPI + 本地量化模型（如 Qwen）**，实现 AI 在小说生成过程中自动抽取关键设定，并写入向量记忆库，从而增强模型的“逻辑一致性”和“记忆保持”。
-
-## 🧩 核心思想
-
-> **AI 自动记忆自己写过的内容，保持小说逻辑统一。**
-
-## 📂 项目结构
-
-（同上，略）
-
-## 🚀 快速启动
+2. Launch the app locally:
 
 ```bash
 bash scripts/launch_local.sh
 ```
 
-默认使用本地 Qwen 模型（量化版本），配置可在 `config.py` 中修改。
-
-## 📌 依赖技术栈
-
-- 🤖 Qwen 本地量化模型（INT4 / bfloat16）
-- 🧱 Langchain 流程构建
-- 🔍 FAISS 向量数据库
-- 🔁 Retrieval-Augmented Generation（RAG）
-- 🛠️ FastAPI 后端接口
-- 🐳 Docker / docker-compose 部署（可选）
+Default setup:
 
-## 📌 TODO（下一阶段）
+- Uses gpt-4o-mini (via OpenAI API) for novel generation
+- Uses bart-large-cnn for chapter summarization  
+- Configurable in config.py (coming soon)
 
-- [ ] 添加多轮生成支持
-- [ ] 实现逻辑评分反馈机制
-- [ ] 云端部署版本（Azure / Hugging Face Spaces）
-- [ ] 单元测试支持
+Requires your own OpenAI API key
 
-## 🧠 项目适用于
+## Coming Soon
 
-- AI + 文学创作方向
-- 多模态 + 长上下文推理实验
-- Langchain + RAG 训练项目
-- 技术型面试展示
+- Chinese-style writing support
+- Multiple books/novels within the same workspace
+- Streaming generation responses to frontend
+- Benchmark consistency improvement with and without memory
+- Deployment to Azure
 
----
+## Contributing
 
-欢迎 Star & Fork，开发中 🚧
+This project is still under active development, Star, Fork, and PRs are welcome!

docker_run.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+docker build -t ai-novelist .
+
+docker run -p 8000:8000 -p 7860:7860 ai-novelist

requirements.txt

@@ -1,6 +1,7 @@
-# model
+# AI
 # torch>=2.0.0
 transformers>=4.36.0
+huggingface_hub
 pillow>=9.0.0
 numpy
 bitsandbytes
@@ -12,14 +13,16 @@ faiss-cpu
 langchain
 langchain-community
 openai
-
-# openai
+pydantic
 einops
 
-
-# fastapipip
+# backend
 fastapi
 uvicorn
 
+# frontend
+gradio
+requests
+
 #jupyter
 ipykernel

scripts/launch_local.sh

@@ -0,0 +1 @@
+uvicorn app.main:app