ohw25_proj_datadashboard_llm

Repo Structure

• contributor_folders temp folder for individual work

• final_notebooks:

  • final_dashboard.py Marimo dashboard
  • llm_working_tutorial demonstration of the working flow of LLM --> plotting
  • functions helper functions of our project
  • txt_docs example functions that agent can use

• scripts backend python scripts to control data access, tools available to the model, etc

• data data will all be cloud accessed. Access can be found in dataset_track.ipynb within this folder.

• photo store figures and photo

• pixi.toml working environment setup

Data Dashboard powered by LLM:

Planning

This reppository hold all code needed to run our LLM-powered climate data analysis dashboard. The project is in its early stages but will consistent of an app interface (containing interactive map, text box, and model output return) as well as a backend for LLM-driven data loading, visualization, and analysis.

• Check out the preliminary code structure here: https://www.figma.com/community/file/1539761856510676831
• Ideation Presentation: Not yet available
• Slack channel: ohw25_proj_llm
• Project google drive: N/A
• Final presentation: Not yet available

Collaborators

Name	Github	Role
Boris Shapkin	boryasbora	Project Facilitator
Liangtong Wei	sryisjelly	Participant
Finn Wimberly	finnwimberly	Participant
Ava Wessel	awessel3	Participant
Aidan Lewis	aidan-axiom	Participant
Dinal Meecle	dinalmeecle	Participant

Background

Boris has some chatbot experience... the rest of us are comfortable with python and excited to learn

Goals

Have a functioning interactive dashboard that users can use natural language ask it to plot the figures they want (e.g., mean sea surface temperature in some region, sea level anomaly time series in a year, etc)

Datasets

can be found here: datasets

Workflow/Roadmap

UI ----->	LLM(Large Language Model) ----->	Create Plot
Use Marimo to build dashboard,	connect llm to one data set for initial testing,	extend as far as we can to achieve the Goals.

Results/Findings

The final marimo dashboard can be found in the final_notebooks folder.

This dashboard comprises of interactive map and a chatbot to select and run analysis on chosen data. All users must have an HF token key from Hugging Face: https://huggingface.co/

The demonstration of LLM working flow in our project can be found in /final_notebooks/llm_working_tutorial folder.

This tutorial provides an exmaple LLM → plotting pipeline. It takes a natural-language plotting request, calls an LLM API to generate pure Python/matplotlib code, and creates the asked plot. It doesn’t train a new LLM, but lets you plug in your own OpenAI-compatible/HF endpoint to generate the plots.

Lessons Learned

Lots of moving parts in this project! Difficult to track keep track of/connect components. Marimo was hyped... but proved difficult. We are not about it. Keep your API keys hidden and protect your pennies.

References

In the works...