Athena is an AI-Assist protoype powered by Cohere-AI and Embed-v3 to faciliate scientific Research. Its key differentiating features include:

• Advanced Semantic Search: Outperforms traditional keyword searches with state-of-the-art embeddings, offering a more nuanced and effective data retrieval experience that understands the complex nature of scientific queries.
• Human-AI Collaboration: Enables easier review of research literature, highlighting key topics, and augmenting human understanding.
• Admin Support: Provides assistance with tasks such as categorization of research articles, e-mail drafting, and tweets generation.

As part of this project we have created two datasets of 50.000 arXiv articles related to AI and NLP using Cohere Embedv3:

• https://huggingface.co/datasets/dcarpintero/arXiv.cs.AI.CL.CV.LG.MA.NE.embedv3
• https://huggingface.co/datasets/dcarpintero/arXiv.cs.CL.embedv3

Steps:

1. Retrieve Articles' Metadata from ArXiv. See ./data_pipeline/retrieve_arxiv.py
2. Embed Articles' Title and Abstract using Embedv3. See ./data_pipeline/embed_arxiv.py
3. Store Articles' Metadata and Embeddings in Weaviate. See ./data_pipeline/index_arxiv.py

Some of our tasks such as enriching abstracts with Wikipedia Links, crafting a glossary, composing e-mails and tweeting rely on a set of:

• Prompt Templates

Those prompts are then composed into a LangChain chain as in the following code snippets:

• Enrich Abstract
• Keywords
• E-mail Drafting w/ JSON Formatting
• Tweet Generation w/ JSON Formatting and Pydantic Validation

See ArxivArticle Class.

The coral.py class provides an abstraction layer over Cohere endpoints.

See app.py

1. Clone the repository:

git@github.com:dcarpintero/athena.git

2. Create and Activate a Virtual Environment:

Windows:

py -m venv .venv
.venv\scripts\activate

macOS/Linux

python3 -m venv .venv
source .venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Run Data Pipeline (optional)

python retrieve_arxiv.py
python embed_arxiv.py
python index_arxiv.py

5. Launch Web Application

streamlit run ./app.py

• Arxiv
• Embed-v3
• Langchain
• Weaviate Vector Search