10minclimate.com

At Boulder Startup Week Builder's Room Uche Ogbuji (Oori Data) had a simple idea: build a simple tool to regularly grab and select climate-related news, and bring these to a subscriber's attention alongside a few relevant action items—so it's not just downer after downer, but a way to feel involved and conscious. Before he knew it, he had a team of 9 folks interested in pitching in, over the course of a 20 hour sprint. We shipped a prototype, came in 2nd place, and 10minclimate.com was born. One inspiration for the project is ClimateCon!, a conference in Boulder, a few weeks after BSW (31 May 2024).

The project continues, and we hope it will continue to be of use to busy people who care about the climate, and need an affirmative way to keep engaged.

Sprint prticipants:

• Micah Dew (LinkedIn)
• Zachariah Malik (LinkedIn)
• Troy Namath (LinkedIn)
• Osi Ogbuji (LinkedIn)
• Uche Ogbuji (LinkedIn)
• Aidan Reese (LinkedIn)
• Garrett Roberts (LinkedIn)
• Elaine Yang (LinkedIn)
• Sung Yi (LinkedIn)

So what does it do, again?

10minclimate.com sends a regular (thrice weekly) e-mail to subscribers (e-mail only at present). This automated e-mail focuses on one digestible, contemporary climate news item, which it summarizes and presents along with three simple actions the reader can take with that news item in mind.

10minclimate.com does run on an LLM workload, and it's well known that AI itself has a Climate Change problem. We ran (courtesy Zachariah Malik (LinkedIn)) some numbers to see that on the M1 Mac Studio we're using to host its LLM inference, the CO₂ contribution is less than that of one person breathing.

Implementation

There are several key components for the project. You can tweak how these are all used in run_daily/config.py

Prerequisites:

Python

Python 3.11 or more recent, preferably in a virtual environment

example setup:

python3.11 -m venv $HOME/.local/venv/bsw
pip install -Ur requirements.txt

SearXNG

Running SearXNG instance. You can just use the Docker container. To run this locally:

export SEARXNG_PORT=8888
docker run --rm \
    -d -p ${SEARXNG_PORT}:8080 \
    -v "${PWD}/searxng:/etc/searxng" \
    -e "BASE_URL=http://localhost:$SEARXNG_PORT/" \
    -e "INSTANCE_NAME=ten-min-climate-engine" \
    searxng/searxng

Note: We want to have some sort of API key, but doesn't seem there is any built-in approach (SEARXNG_SECRET is something different). We might have to use a reverse proxy with HTTP auth.

This gets SearXNG runing on port 8888. Feel free to adjust as necessary in the 10minclimate.com config.

You do need to edit searxng/settings.yml relative to where you launched the docker comtainer, making sure server.limiter is set to false and - json is included in search.formats.

You can then just restart the continer (use docker ps to get the ID, docker stop [ID] and then repeat the docker run command above).

Running on a shared server

For production SearXNG will need to run on a shared server. Make sure that server has Docker installed, set it as the context. Create /etc/searxng on the remote server and ensure it's writable by the docker daemon.

sudo chgrp docker /etc/searxng
sudo chmod g+ws /etc/searxng

Then launch with the following setup, where /etc/searxng is mounted

export SEARXNG_PORT=8888
docker run --rm \
    -d -p ${SEARXNG_PORT}:8080 \
    -v "/etc/searxng:/etc/searxng" \
    -e "BASE_URL=http://localhost:$SEARXNG_PORT/" \
    -e "INSTANCE_NAME=oorihive-engine" \
    searxng/searxng

LLM endpoint(s)

Uses llama.cpp remotely hosted for LLM processing.

Set up an endpoint, and update your environment

3rd-party python libraries

From your virtual environment:

pip install -Ur requirements.txt

Running the daily command

run_daily/main.py will:

• pull news for the past day
• run LLM-based assessments, summarization, credibility scoring & storage of best candidates for next action e-mail
• check e-mail day criteria (as configured, e.g. Tuesday, Thursday and Saturdays)
• if it's an e-mail day, pull all pending news item candidates & LLM-generate action items
• Send e-mail to gethered addresses

Until you're sure you know what you're doing, you probably want to use the dry run feature (see testing section, below)

Testing

Setting environment variables:

Use a sound approach to set up the environment, including secrets.

For example, if you use 1password and have a file op.env, you can just set the environment inline to run any COMMAND:

op run —env-file op.env -- COMMAND

Test command:

So in the case of tsting:

dotenv -f op.env run -- op run --env-file=op.env -- run_daily/main.py --dry-run

example .env

SEARXNG_ENDPOINT="http://localhost:8888/search"
SEARXNG_LIMIT="6"

SUMMARIZATION_LLM_URL="http://localhost:8000"
SCORING_LLM_URL="http://localhost:8000"
ACTION_GEN_LLM_URL="http://localhost:8000"

CLIMATE_ACTION_DB_NAME="climateDB"
CLIMATE_ACTION_DB_HOST="localhost"
CLIMATE_ACTION_DB_PORT="1234"
CLIMATE_ACTION_DB_USER="user"
CLIMATE_ACTION_DB_PASSWORD="password"

MAILCHIMP_API_KEY="key"
MAILCHIMP_API_SERVER="localhost"
MAILCHIMP_AUDIENCE_ID="12345"

You can use both --dry-run and --set-date specify a date to test how the program would operate on a given day.

run_daily/main.py --dry-run --set-date=2024-05-23

Docker setup

❯ docker build -t tmclimate-cron -f ./docker/rundaily/dockerfile  

Design notes

• Separates language from code using Word Loom

TODO

Product

• Personalization, which will be fed into prompting for summarization & suggested actions
• Credibility checking of news items

Dev

• Data flow & other engineering diagrams

• Proper library setup & modularization (now all spaghetti-thrown into the run_daily dir)

• Online archive of past action e-mails

• Continue to think about managing/securing SearXNG (as well as PGVector & llama.cpp). Security-minded "Searx Installation and Discussion" article.

HISTORY: WHITEBOARD notes from Kickoff day (May 14)

Problem:

keeping up with climate news is overwhelming and demoralizing

Solution:

• Hub to push info/ideas/actions to user based on climate news
• utilizing AI to customize/agregate/summarize(/score on the backend?) feed based on specific interests
• email the user actionable steps
  • daily weather
  • summarize relevant news (for the user)
  • give users a selection of action items relevant to the situation

maybe out of scope, but cool ideas!

• gamification?
  • UX focused around a big thermometer that's goin up with climate change and when you do an action, you make a dent on it!
• show/record what the user has done to keep them invested
• (geographical) scale filtering?
• politicization?
• group action?