DocChat (Production RAG Pipeline)

Upload .txt or .pdf documents, ask questions in plain English.

Live Demo → basic-rag-mstghx3dbkhezdategemlv.streamlit.app

---

Evaluation Results (RAGAS)

Evaluated on a QA dataset over the included articles (python evaluate.py):

Metric	Score	What it means
Faithfulness	0.92	Answers grounded in retrieved context
Answer Relevancy	0.88	Answers address the question asked
Context Precision	0.85	Retrieved chunks are on-topic
Context Recall	0.81	Necessary chunks were retrieved

Run python evaluate.py to reproduce with your own documents.

---

Architecture

User Question
      │
      ▼
┌─────────────────────────────────────────┐
│       Streamlit UI  /  CLI (main.py)    │
└──────────────────┬──────────────────────┘
                   │
      ┌────────────┴────────────┐
      ▼                         ▼
┌──────────────┐      ┌──────────────────────┐
│ data_loader  │      │    vector_store      │
│              │      │                      │
│ .txt + .pdf  ├─────▶│  OpenAI Embeddings   │
│ LangChain    │chunks│ ChromaDB (persisted) │
│ TextSplitter │      │  Similarity search   │
└──────────────┘      └──────────┬───────────┘
                                 │ top-k chunks
                                 ▼
                      ┌──────────────────────┐
                      │       rag.py         │
                      │                      │
                      │ LangChain LCEL chain │
                      │  GPT-4o-mini         │
                      │  Source citations    │
                      └──────────┬───────────┘
                                 │
                                 ▼
                      ┌──────────────────────┐
                      │     RAGResponse      │
                      │  answer: str         │
                      │  sources: list[str]  │
                      │ num_chunks_used: int │
                      └──────────────────────┘

---

Quick Start

1. Clone and install

git clone https://github.com/Babarali2k21/basic-rag.git
cd basic-rag

python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

pip install -r requirements.txt

2. Configure

cp .env.example .env
# Edit .env and set OPENAI_API_KEY=sk-your-key-here

3. Add your documents

Drop .txt or .pdf files into the ./articles/ folder. A sample article is included to get you started.

4a. Web UI (recommended)

streamlit run app.py

Open http://localhost:8501, enter your API key in the sidebar, upload documents or use the articles folder, then start asking questions.

4b. CLI

# Index documents
python main.py --index

# Interactive Q&A session
python main.py

# Single question
python main.py --query "What is retrieval-augmented generation?"

5. Run evaluation (optional)

python evaluate.py

---

Project Structure

basic-rag/
├── app.py              # Streamlit web UI with document upload
├── main.py             # CLI (--index / --query / interactive)
├── config.py           # Settings via Pydantic BaseSettings
├── data_loader.py      # Load .txt/.pdf + LangChain text splitting
├── vector_store.py     # ChromaDB indexing, loading, retriever factory
├── rag.py              # LangChain LCEL chain + RAGResponse dataclass
├── evaluate.py         # RAGAS evaluation script
├── conftest.py         # pytest path configuration
├── articles/           # Drop knowledge base files here
│   └── rag_overview.txt
├── docs/
│   └── screenshot.png  # App screenshot
├── tests/
│   └── test_rag.py     # Unit + integration tests (mocked, no API key needed)
├── .env.example        # Environment variable template
├── requirements.txt
└── .github/
    └── workflows/
        └── tests.yml   # CI: pytest + codecov on Python 3.11 & 3.12

---

Configuration

All settings via .env file:

Variable	Default	Description
OPENAI_API_KEY	—	Required
OPENAI_MODEL	gpt-4o-mini	LLM for answer generation
EMBEDDING_MODEL	text-embedding-3-small	Embedding model
CHUNK_SIZE	512	Characters per chunk
CHUNK_OVERLAP	64	Overlap between chunks
RETRIEVAL_K	5	Chunks to retrieve per query
CHROMA_PATH	chroma_persistent_storage	ChromaDB persistence path
ARTICLES_DIR	./articles	Knowledge base directory

---

Tests

pytest tests/ -v

# With coverage report
pytest tests/ -v --cov=. --cov-report=term-missing

All tests use mocked LLM calls no API key required to run the test suite.

12/12 tests passing on Python 3.11 and 3.12.

---

Roadmap

• Core RAG pipeline (load → chunk → embed → retrieve → generate)
• PDF support via PyPDF
• Document upload via Streamlit UI
• LangChain LCEL chain with source citations
• Streamlit web UI with chat history
• RAGAS evaluation script
• Pydantic v2 settings
• 12 pytest tests (mocked, no API key needed)
• CI/CD (GitHub Actions + Codecov)
• Deployed on Streamlit Cloud
• Conversational memory (multi-turn Q&A)
• Advanced retrieval: HyDE + cross-encoder reranking
• Pinecone / Qdrant as vector store alternatives
• Docker deployment

---

Tech Stack

Layer	Tool
LLM	OpenAI GPT-4o-mini
Embeddings	OpenAI text-embedding-3-small
RAG Framework	LangChain 0.2
Vector Store	ChromaDB
Evaluation	RAGAS
Web UI	Streamlit
Config	Pydantic BaseSettings
Testing	pytest + pytest-cov (12 tests)
CI/CD	GitHub Actions + Codecov
Deployment	Streamlit Cloud

---

Author

Babar Ali AI Engineer · Vienna, Austria

---

License

MIT