DeepSeek Danger Detection

This repository includes a sample analysis_tool that:

1. Reads a YOLO detection log file (detections.txt).
2. Invokes a DeepSeek model via Ollama to analyze the log.
3. Produces an email-style summary with the number of Danger detections, a short explanation, an optional warning, and a mention of a detection_snapshot.png.
4. Writes the final output to analysis.txt.

How it Works

1. analysis_tool (defined in deepseek_agent.py):

   • Loads the YOLO detections from a file.
   • Sends a prompt to the DeepSeek model via Ollama.
   • Parses the AI response, ensuring we have bullet lines for:
     • - Number of 'Danger' detections: X
     • - Brief explanation: Y
   • Adds some fallback text if the model doesn't provide them.
   • Saves the final email-like text to analysis.txt.

2. Mail Sending (Future Step):

   • Another agent (or Python code) can read analysis.txt and send it to the required email address.

Requirements

• Python 3.8+
• langchain-ollama for ChatOllama
• Ollama CLI or server running locally with your chosen model (e.g. deepseek-r1:1.5b)
• Additional packages as listed in requirements.txt.

Usage

1. Make sure you have Ollama installed and running locally.
2. Pull the model:

   ollama pull deepseek-r1:1.5b
   

ollama serve

from deepseek_agent import analysis_tool

result = analysis_tool('detections.txt') print(result)