DynPrompt-1s

Dynamic Prompt Assistant - Single Agent with Streamlit UI

Key Features

Your free tool for completing repetitive tasks with generative AI

$0 subscription fee. 100% data privacy.

About the Project

PyCon Hong Kong 2024 Sprint project.

This Python app utilises gen AI for repetitive, similar tasks. This allows users to input task details into a CSV format, efficiently delegating these responsibilities to a locally hosted large language model powered by Ollama. The system then generates responses and returns them in a readily usable CSV format.

Technical Details

• Python version: 3.11
• Libraries used: ollama-python
• Dependencies required: pandas, streamlit

Installation and Usage

How to Use the Tool:

1. Prepare Your Data: Ensure your data is in a single-column CSV format. Reformat your data as needed to meet this requirement by following the guidelines.

2. Upload Your File: Click the Upload button to upload your formatted file. A preview of a portion of your data will be displayed for review.

3. Enter Your Command: In the Prompt Command text field, specify the task you want the tool to perform with the uploaded data (e.g., analyze, manipulate).

4. Run the Task: Click the Run button to execute the task.

5. Download Results: Once the task is completed, click the Download button to download the results as a CSV file.

Download and install Ollama from https://ollama.com/download

In Terminal:

pip install -r requirements.txt
ollama pull llama3.2:3b

Prep data.csv

prompt_var is the content of your task.

Paste each of them in the prompt_var column in separate rows.

Set your prompt

Edit your prompt under Base prompt section in ui.py. Use prompt engineering techniques for desired results.

Example use cases:

• Text extractor for email

  prompt = f"""
      Given is an email from contains product code and corresponding remarks. Identify the product code and remarks within the original text. 
      
      Provide the product code and remark only in csv format, ready to save. Exclude the "```csv" declaration, don't add spaces after the comma, include column headers.
  
      Format:
      product_code, remark
      product_code_1, remark_1
      ...
      
      Email:
      """
  

• Customer review text classifier

  prompt = f"""
      Respond with "Positve" or "Negative": 
      the comment is a product reivew, describing an user experience. 
  
      Comment:
      """
  

Run ui.py

cd DynPrompt-1s
streamlit run ui.py 

Results in responses.csv

responses.csv is saved in the same folder. Open in Excel.

Challenges and Limitations

The responses are largely depend on the LLM in use, and prompt engineering techniques.

Prompt engineering techniques can be refined with resources mentioned above.

What LLM model could be used depends on your hardware. The better the LLM model, the higher RAM and VRAM requirements.

In this script in particular, I am using llama3.2:3b on MacBook M3 Pro 18GB RAM. You may run into errors if your VRAM is insufficient.

Check in terminal to see if Ollama is running correctly. Or edit the script for a smaller LLM accordingly.

ollama run llama3.2:3b