Generative AI Assistant Tool on Minerva

• Open WebUI Header & Sidebar Toggle ([|]): Expand or collapse the sidebar menu.
• New Chat: Click to clear the current workspace and start a fresh session.
• Search: Search through your previous prompt history and saved chats.
• Notes: Access and organize saved workspace notes.
• Folders & Chats: Organize your active conversations into dedicated folders or access saved individual chat sessions.

• Model Indicator: Displays the active model currently selected for the conversation (e.g., OI llama4:latest).
• Prompt Input Box:
  • Add Files (+): Upload documents (PDFs, text files) or images to interact with using Retrieval-Augmented Generation (RAG) or multimodal vision tasks.
  • Tools/Plugins (❖): Access integrations, web tools, or advanced interpreter tools.
  • Model Switcher Dropdown: Click the dropdown next to the model name inside the prompt box (e.g., llama4:latest ˅) to switch between supported models running on Minerva HPC.
  • Voice Input (🎙): Dictate your prompt using microphone input.
  • Voice Response Mode (|||): Toggle interactive audio mode.
• Suggested Starters: Quick-click suggestions to quickly test model capabilities (e.g., study help, brainstorming, or productivity tips).

• Controls Icon (⚙ / Controls): Adjust session-specific model parameters such as temperature, system prompt, or context length.
• User Profile & Settings: Access account settings, documentation, keyboard shortcuts, or sign out.

Ask target scientific questions directly via the prompt bar. You can follow up with suggested prompts or type custom questions to explore topics interactively.

Example Output Focus: Summarizes aspects like precision/efficiency, ease of use, specificity, versatility across organisms, and reduced off-target effects compared to TALENs or ZFNs.

Paper Summarization — Upload documents (e.g., PDFs) by clicking the + icon in the input panel and selecting Upload Files.

Example Output: Extracts key methodology, framework details (e.g., federated learning across multiple hospitals), evaluation metrics (AUROC values), guidelines followed (TRIPOD), and license info.

Conversational Document Interaction — Ask follow-up questions specific to the uploaded file’s contents.

Example Output: RAG retrieves contextual details from the paper, explaining that the federated model outperforms local models (especially in smaller datasets) and performs comparably to the pooled model.

Translates complex clinical terminology into clear, accessible descriptions.

Example Output: Defines AKI as a sudden loss of kidney function over hours or days, provides intuitive analogies (e.g., a coffee filter), lists potential causes (dehydration, NSAIDs/antibiotics, sepsis) and common symptoms (fatigue, swelling, altered urination patterns).

Get help writing, debugging, or understanding code in multiple languages.

import numpy as np
from scipy import stats

# Sample data
np.random.seed(0)
sample1 = np.random.normal(0, 1, 100)
sample2 = np.random.normal(0.5, 1, 100)

# Perform two-sample t-test
t_stat, p_val = stats.ttest_ind(sample1, sample2)

print(f"T-Statistic: {t_stat}")
print(f"P-Value: {p_val}")

# Interpret the results
alpha = 0.05
if p_val < alpha:
    print("Reject the null hypothesis. The means are likely different.")
else:
    print("Fail to reject the null hypothesis. The means are likely the same.")

Translate clinical notes, queries, or descriptions into various target languages.

Generated Response: “¿Cómo te sientes hoy? Revisar los síntomas físicos como dolor, fatiga o molestias puede ayudar a detectar problemas de salud temprano y guiar el cuidado adecuado.”

Evaluate accuracy, response depth, and clarity across different model sizes concurrently.

How to enable: Click the + icon next to the active model name in the chat panel to add additional models (e.g., compare llama4:latest, llama3.3:70b, and tinyllama:latest).

Result: Displays side-by-side columns with each model’s response generated simultaneously from a single prompt.

Integrate the Assistant Tool directly into custom Python scripts, data pipelines, or applications.

1. Obtain an API key by navigating to Settings > Account in the web interface.
2. Use the endpoint URL: https://assistant.hpc.mssm.edu/api/chat/completions

Python API Example:

import requests

token = "your_api_token_here"
url = 'https://assistant.hpc.mssm.edu/api/chat/completions'

headers = {
    'Authorization': f'Bearer {token}',
    'Content-Type': 'application/json'
}

data = {
    "model": "tinyllama:latest",
    "messages": [
        {
            "role": "user",
            "content": "What are the main causes of cardiovascular disease?"
        }
    ]
}

response = requests.post(url, headers=headers, json=data)
result = response.json()

if 'choices' in result and result['choices']:
    print("Assistant response:", result['choices'][0]['message']['content'])
else:
    print("No assistant response found.")