Run data analysis with a Python agent on a local model
A commenter says LangChain's create_python_agent works reasonably well with the Nous Hermes model for data analysis.
Done with a custom-built or unnamed agent
- What they did
- The thread is very thin: the original poster asked about using Llama 2 with LangChain to analyze a CSV, after finding LangChain’s pandas-capable Python agent worked well with OpenAI models but poorly with a quantized 7B Llama. In replies, the same user said to use LangChain’s create_python_agent instead of the pandas agent, and said it worked reasonably well with the nous-hermes-llama2 13B model.
- How it went
- The only reported outcome is qualitative: the quantized 7B Llama gave disappointing results, while create_python_agent with Hermes/Nous Llama 2 13B worked reasonably well. The thread gives no example task, metrics, or failure details.
- Worth knowing
- The main catch is that the source does not include a workflow, prompt, or benchmark; it only names create_python_agent and the Nous Hermes Llama 2 13B model.
Try it yourself with Hermes Agent
I have a dataset at [path to CSV or data file] and I want to understand [question, e.g. which products drive the most revenue by month]. Load the file, check for missing or odd values, then run the analysis by writing and executing Python code, showing me the code and results at each step. Finish with a short plain-language summary of the findings and one chart saved to [output folder]. Do not modify or delete the original data file.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · Aug 5, 2023
Related use cases
Find connections across 100 non-fiction books
Claude Code browses a personal library with CLI tools and surfaces trails of related excerpts across books.
Build searchable knowledge bases from videos and papers
An OpenClaw agent runs a YouTube summariser and a paper-condensing tool, saving Markdown into two indexed knowledge bases for videos and papers.
Turn voice notes into structured Obsidian notes
Voice notes sent over Telegram become Obsidian thought notes with transcript, cleaned argument, linked context and a draft; Hermes also tracks recipes.
Monitor tech subreddits and competitors, draft emails and articles
Commenters use Cowork daily to scan Reddit tech subreddits for work items, monitor competitors and markets, generate articles and draft emails.
Track competitor prices and contacts into Google Sheets
A Claude-powered scraper agent monitors competitor sites hourly, pulling pricing, company details and contacts into Google Sheets, replacing manual research.
Run a weekly competitive-monitoring briefing for product management
A product manager had Hermes produce a weekly competitor briefing; its skill rewrote itself four times and by week six surfaced patterns he had missed manually.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.