# rag-gemini-multi-modal

Multi-modal LLMs enable visual assistants that can perform question-answering about images. 

This template create a visual assistant for slide decks, which often contain visuals such as graphs or figures.

It uses OpenCLIP embeddings to embed all of the slide images and stores them in Chroma.
 
Given a question, relevat slides are retrieved and passed to [Google Gemini](https://deepmind.google/technologies/gemini/#introduction) for answer synthesis.
 
![Diagram illustrating the process of a visual assistant using multi-modal LLM, from slide deck images to OpenCLIP embedding, retrieval, and synthesis with Google Gemini, resulting in an answer.](https://github.com/langchain-ai/langchain/assets/122662504/b9e69bef-d687-4ecf-a599-937e559d5184 "Workflow Diagram for Visual Assistant Using Multi-modal LLM")

## Input

Supply a slide deck as pdf in the `/docs` directory. 

By default, this template has a slide deck about Q3 earnings from DataDog, a public techologyy company.

Example questions to ask can be:
```
How many customers does Datadog have?
What is Datadog platform % Y/Y growth in FY20, FY21, and FY22?
```

To create an index of the slide deck, run:
```
poetry install
python ingest.py
```

## Storage

This template will use [OpenCLIP](https://github.com/mlfoundations/open_clip) multi-modal embeddings to embed the images.

You can select different embedding model options (see results [here](https://github.com/mlfoundations/open_clip/blob/main/docs/openclip_results.csv)).

The first time you run the app, it will automatically download the multimodal embedding model.

By default, LangChain will use an embedding model with moderate performance but lower memory requirments, `ViT-H-14`.

You can choose alternative `OpenCLIPEmbeddings` models in `rag_chroma_multi_modal/ingest.py`:
```
vectorstore_mmembd = Chroma(
    collection_name="multi-modal-rag",
    persist_directory=str(re_vectorstore_path),
    embedding_function=OpenCLIPEmbeddings(
        model_name="ViT-H-14", checkpoint="laion2b_s32b_b79k"
    ),
)
```

## LLM

The app will retrieve images using multi-modal embeddings, and pass them to Google Gemini.

## Environment Setup

Set your `GOOGLE_API_KEY` environment variable in order to access Gemini.

## Usage

To use this package, you should first have the LangChain CLI installed:

```shell
pip install -U langchain-cli
```

To create a new LangChain project and install this as the only package, you can do:

```shell
langchain app new my-app --package rag-gemini-multi-modal
```

If you want to add this to an existing project, you can just run:

```shell
langchain app add rag-gemini-multi-modal
```

And add the following code to your `server.py` file:
```python
from rag_gemini_multi_modal import chain as rag_gemini_multi_modal_chain

add_routes(app, rag_gemini_multi_modal_chain, path="/rag-gemini-multi-modal")
```

(Optional) Let's now configure LangSmith. 
LangSmith will help us trace, monitor and debug LangChain applications. 
LangSmith is currently in private beta, you can sign up [here](https://smith.langchain.com/). 
If you don't have access, you can skip this section

```shell
export LANGCHAIN_TRACING_V2=true
export LANGCHAIN_API_KEY=<your-api-key>
export LANGCHAIN_PROJECT=<your-project>  # if not specified, defaults to "default"
```

If you are inside this directory, then you can spin up a LangServe instance directly by:

```shell
langchain serve
```

This will start the FastAPI app with a server is running locally at 
[http://localhost:8000](http://localhost:8000)

We can see all templates at [http://127.0.0.1:8000/docs](http://127.0.0.1:8000/docs)
We can access the playground at [http://127.0.0.1:8000/rag-gemini-multi-modal/playground](http://127.0.0.1:8000/rag-gemini-multi-modal/playground)  

We can access the template from code with:

```python
from langserve.client import RemoteRunnable

runnable = RemoteRunnable("http://localhost:8000/rag-gemini-multi-modal")
```