Skip to content

Running the Streamlit Application

This guide walks through running the AI Research Assistant locally—from cloning the repository to asking your first research question.

Note

This project is designed for Apple Silicon Macs and uses MLX-LM for fully local large language model inference.

1. Fork the Repository

Navigate to the repository:

https://github.com/Cristy210/AI_RA

Click Fork in the upper-right corner of GitHub to create your own copy.

2. Clone Your Fork

git clone https://github.com/<your-github-username>/AI_RA.git

cd AI_RA

3. Create the Conda Environment

The repository includes an environment.yml file containing all required dependencies.

Create the environment:

conda env create -f environment.yml

Activate it:

conda activate AIRA

4. Launch the Streamlit Application

From the project root, start Streamlit:

streamlit run app.py

Streamlit will start a local web server.

Open the URL displayed in your terminal (typically http://localhost:8501).

5. Wait for the Language Model

When the application starts for the first time, it loads the local MLX language model:

mlx-community/Mistral-7B-Instruct-v0.3-4bit

The initial launch may take a few minutes while the model is downloaded and cached.

Subsequent launches are significantly faster.

6. Create Your First Research Database

When the application opens, the sidebar displays two options:

  • Create a new database
  • Select an existing database

Choose Create a new database.

Provide:

Field Example
Database Name Subspace_Clustering
arXiv Research Topic subspace clustering
Maximum Papers 10

Click Build Database.

The application will automatically:

  1. Search arXiv
  2. Download matching papers
  3. Parse PDFs
  4. Split documents into text chunks
  5. Generate embeddings using
BAAI/bge-small-en-v1.5
  1. Store the embeddings in a persistent Chroma vector database
  2. Register the database for future use

Depending on the number of papers selected, this process may take several minutes.

7. Ask Your First Question

Once indexing completes, the selected database becomes active.

Use the chat input at the bottom of the page to ask a research question, for example:

What optimization methods are used in subspace clustering algorithms?

The application will:

  1. Retrieve the most relevant document chunks.
  2. Construct a grounded prompt.
  3. Generate an evidence-based response using the local MLX model.

Working with Multiple Databases

The application supports multiple research databases.

Use the sidebar to switch between databases without rebuilding embeddings.

Each database stores:

  • Research topic
  • Number of indexed papers
  • Number of document chunks

Selecting a different database automatically loads its corresponding Chroma collection.

Deleting a Research Database

The Streamlit application allows you to permanently remove a research database when it is no longer needed.

To delete a database:

  1. Select an existing database from the sidebar.
  2. Expand Delete database.
  3. Type the database name exactly as displayed to confirm the deletion.
  4. Click Delete Database.

Reusing Existing Databases

Previously created databases are stored locally and automatically detected the next time the application starts.

Simply launch Streamlit again:

conda activate AIRA

streamlit run app.py

Then select the desired database from the sidebar.

No re-indexing is required unless you want to create a new research database.

Creating a Different Knowledge Base

To build a database for another research area:

  1. Select Create a new database
  2. Enter a new database name
  3. Enter a different arXiv search topic
  4. Click Build Database

For example:

Database Name Research Topic
julia_ml Julia adaptation in machine learning
LLMs large language models
graph_clustering graph clustering

Each topic is indexed independently and can be selected from the sidebar whenever needed.