🔬 Research Assistant | 研究助手

An AI-powered research assistant with RAG, LoRA fine-tuning, and multi-model support
基于RAG检索增强和LoRA微调的智能研究助手

---

✨ Features | 功能特性

---

🏗️ Architecture | 架构

research_assistent/
├── config/               # YAML & JSON configuration
├── src/
│   ├── llm/              # OllamaClient + ModelManager
│   ├── rag/              # Document loading, embeddings, vector store, retriever
│   ├── lora/             # LoRA fine-tuning with PEFT
│   ├── memory/           # Conversation DB + Knowledge Graph
│   └── utils/            # PDF processor, text utils, logger
├── ui/
│   ├── cli.py            # Click-based CLI
│   └── web.py            # Streamlit web app
├── scripts/              # Setup and utility scripts
├── tests/                # pytest test suite
├── data/                 # Papers, ChromaDB, LoRA adapters
└── Dockerfile + docker-compose.yml

Data Flow:

PDF/Text → DocumentLoader → Chunker → EmbeddingManager → VectorStore (ChromaDB)
                                                               ↓
User Query → OllamaClient ←── Retriever ←── similarity_search()
                ↓
         RAG Response + Sources

---

🚀 Quick Start | 快速开始

Option A: Local Setup

# 1. Clone and setup
git clone https://github.com/example/research_assistent.git
cd research_assistent
bash scripts/setup.sh
source venv/bin/activate

# 2. Start Ollama and pull a model
curl -fsSL https://ollama.ai/install.sh | sh
ollama serve &
ollama pull llama2

# 3. Launch the Web UI
streamlit run ui/web.py
# OR use the CLI
python ui/cli.py --help

Option B: Docker

docker compose up -d
# Open http://localhost:8501

---

📖 CLI Usage | 命令行使用

# Interactive chat
python ui/cli.py chat --model llama2 --domain machine_learning

# Review a paper
python ui/cli.py review paper.pdf

# Add paper to knowledge base
python ui/cli.py add-paper paper.pdf --domain machine_learning

# Writing assistance
python ui/cli.py write --type abstract --topic "LoRA fine-tuning"

# Fine-tune with your data
python ui/cli.py train --data training_data.json --epochs 3

---

🐍 Python API

from src.llm.ollama_client import OllamaClient
from src.rag.document_loader import DocumentLoader
from src.rag.embeddings import EmbeddingManager
from src.rag.vector_store import VectorStore
from src.rag.retriever import Retriever

client = OllamaClient(model="llama2")
loader = DocumentLoader(chunk_size=1000, chunk_overlap=200)
emb    = EmbeddingManager()
store  = VectorStore(persist_directory="./data/chroma_db", embedding_manager=emb)

docs = loader.load_and_split("paper.pdf")
store.add_documents(docs)

retriever = Retriever(store, client)
result = retriever.rag_query("What is the main contribution?")
print(result["answer"])

---

⚙️ Configuration | 配置

Edit config/config.yaml to change model, RAG settings, etc.

llm:
  model: "llama2"
  temperature: 0.7
  max_tokens: 2048

rag:
  chunk_size: 1000
  top_k: 5
  embedding_model: "sentence-transformers/all-MiniLM-L6-v2"

---

🔧 Development

pytest tests/ -v
pip install -e ".[dev]"

---

🤝 Contributing

1. Fork → feature branch → changes + tests → PR

📄 License

MIT License

🙏 Acknowledgements

Ollama · LangChain · ChromaDB · PEFT · Streamlit