https://github.com/unikill066/agentic-rag
This isn't just another RAG implementation, it's an agentic system that intelligently decides when and how to retrieve information.
https://github.com/unikill066/agentic-rag
agentic-ai agentic-rag langchain langchain-python langgraph langgraph-python rag
Last synced: 14 days ago
JSON representation
This isn't just another RAG implementation, it's an agentic system that intelligently decides when and how to retrieve information.
- Host: GitHub
- URL: https://github.com/unikill066/agentic-rag
- Owner: unikill066
- License: mit
- Created: 2025-06-22T15:10:55.000Z (10 months ago)
- Default Branch: main
- Last Pushed: 2025-07-04T20:05:13.000Z (10 months ago)
- Last Synced: 2025-07-04T21:20:16.298Z (10 months ago)
- Topics: agentic-ai, agentic-rag, langchain, langchain-python, langgraph, langgraph-python, rag
- Language: Python
- Homepage: https://agentic-rag-gnqmsrmdhun5d9f4jua9p2.streamlit.app/
- Size: 9.23 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Agentic RAG Chatbot
[](LICENSE)
An **agentic Retrieval-Augmented Generation (RAG)** chatbot powered by [LangGraph](https://github.com/unikill066/langgraph), OpenAI’s GPT-3.5-turbo, and a Chroma vector store. It allows you to ask questions about **Nikhil Nageshwar Inturi’s** background, publications, projects, and qualifications, and get grounded answers sourced from indexed PDFs and other documents.
The app supports one upload at a time; progress bars updates live. More on the steps that go into the pipeline can be found here:
```mermaid
graph TD
A[Start] --> B[RAG Agent]
B --> C{tools_condition}
C -->|tools| D[Retriever Node]
C -->|END| E[End]
D --> F{document_quality}
F -->|generator| G[Generator]
F -->|rewrite| H[Rewrite]
G --> E
H --> B
```
### Difference b/w [Traditional RAG](https://nikhilrag-ntbaxj9puvp37yaqvkqsiu.streamlit.app/) & Agentic RAG
| Feature | Traditional RAG | Agentic RAG |
|------------------------------------|---------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------|
| **Overview** | Retriever → Generator; static pipeline grounded in retrieved context. | RAG + intelligent agent; dynamic decision-making, planning & tool use. |
| **Architecture** | Fixed sequence: query → embedding → retrieve top-k → prompt → LLM answer. | Loop of “Thought → Tool → Observation”; agent decides when/where/how to retrieve or call tools. |
| **Workflow** | One-shot: embed & retrieve → augment prompt → generate answer. | Iterative: plan → retrieve → evaluate → (re-retrieve…) → generate → (optional verify). |
| **Memory & Reasoning** | No built-in memory; reasoning is one-pass inside the final LLM call. | Maintains state across steps; chain-of-thought exposes planning, self-correction & long-term memory. |
| **Use Cases** | Simple QA on a single knowledge source. | Complex, multi-step tasks; multi-source routing; tool-augmented actions (e.g. charts, APIs). |
| **Trade-offs** | Simpler, faster, cheaper, easier to debug; limited adaptability. | More flexible & accurate; handles ambiguity; higher latency, cost & engineering complexity. |
| **LangChain | Chains: static RetrievalQA/ConversationRetrieval. | Agents: wrap retrievers as tools (LangChain Agents) |
> Note: Check out the RAG implementation in this repository: [NikhilRAG](https://github.com/unikill066/NikhilRAG)
## Features
* **Agentic RAG pipeline**
A state‐graph (`graph.py`) orchestrates:
1. A router/agent node that decides whether to call a document retriever tool
2. A document‐quality checker to route between rewriting or generation
3. A generator node that synthesizes answers from retrieved context
4. A rewrite node that reformulates queries when no relevant docs are found
* **Custom retriever tool**
Uses [Chroma](https://github.com/langchain-community/langchain-community) to index and retrieve document chunks (via `embed_generator.py` → `./chroma_db`), exposed as a named tool for the agent.
* **Streamlit front-end**
A chat UI (`streamlit_app.py`) to interact with your RAG agent in real time:
* Initializes and persists `st.session_state.messages`
* Shows a welcome prompt and conversation history
* Invokes the compiled LangGraph state‐graph under the hood
* Displays “Thinking…” spinners and debug info on demand
* **Embeddings & indexing**
`embed_generator.py` loads PDFs from `./docs`, splits into chunks, and persists embeddings in `./chroma_db`.
* **Retrieval QA demo**
A standalone retrieval‐QA script (`bin/retriever.py`) to sanity-check your vector store outside of the state graph.
## Getting Started
### 1. Clone the repo
```bash
git clone https://github.com/unikill066/agentic-rag.git
cd agentic-rag
```
### 2. Install dependencies
Using **pip** (via `requirements.txt`):
```bash
pip install uv
uv venv .venv
source .venv/bin/activate
uv pip install -r requirements.txt
```
Or with **Poetry**, `uv` is just 10-100X faster (via `pyproject.toml`):
```bash
poetry install
poetry shell
```
### 3. Environment variables
Create a `.env`(refer to .env.example) in the repo root and set your OpenAI API key (and, if used, Firebase credentials):
```ini
OPENAI_API_KEY=sk-...
LANGSMITH_TRACING=true
LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
LANGSMITH_API_KEY=ls-...
LANGSMITH_PROJECT="proj_name"
```
### 4. Index your documents
Put any `.pdf` files you want to query into `./docs/` then run:
```bash
python bin/embed_generator.py
```
This will split and embed your PDFs into `./chroma_db`.
### 5. Run the Streamlit app
```bash
streamlit run streamlit_app-local.py
```
Open your browser at `http://localhost:8501` and start asking questions!
## Repo Structure
```
.
├── bin/
│ ├── embed_generator.py # PDF -> Chroma embedding pipeline
│ └── retriever.py # Standalone RetrievalQA demo for testing
├── chroma_db
├── constants.py
├── graph.py # builds & compiles the LangGraph Agentic RAG agent
├── streamlit_app.py # Streamlit front‐end & chat UI
├── requirements.txt
├── pyproject.toml
└── README.md
```
## License
This project is licensed under the [MIT License](LICENSE).
> Built by [Nikhil Nageshwar Inturi](https://github.com/unikill066) • 2025-06-22
>
> Contact: [ Gmail ](mailto:inturinikhilnageshwar@gmail.com)