Skip to content

Designing my AI nanny share bot

Published: at 10:20 PM

Yesterday, I decided to try to build an app to help me filter through the many nanny- and nanny-share-related posts in local Facebook groups.

To kick off this project, I’ll first do some design:

Desired app behavior

The primary thing I want to automate is sifting through the posts in my various Facebook groups, and picking out the ones I should take a closer look at.

As a V1:

Flow of data

StepWhat happensData it needsData it outputs
Step 1: Gather postsOpen up each of my Facebook groups, and get the new posts.- URLs of Facebook groups (and my logged-in state)
- Timestamp of the last time we gathered posts
- New posts
Step 2: Filter posts1. Look at each of the new posts.
2. “Understand” each post and determine if it matches the criteria I want.
3. If yes, “mark” the post as interesting.
- New posts
- Criteria I want
- A “yes/no” decision of whether each new post is interesting to me
Step 3: Create summary1. Aggregate the posts “marked” as interesting.
2. Write a summary that features the text of the posts.
- New posts marked as interesting- A summary

Additional design goals & constraints

Beyond the basic functionality of my app, there are two things I want to design for.

Data privacy

Ideally, I’d like to not send the Facebook group data to third parties like OpenAI (creator of ChatGPT) or Anthropic (creator of Claude).

Thus, I’ll aim to use an open source LLM that I can self-host. I can run an LLM on my laptop, so the data never leaves my computer.

Terms of service

Facebook’s terms of service don’t allow using a bot to download their data.

So I’ll plan to do “Step 1: Gather posts” manually first. (However, I’ll build what tools I can to make this step as quick as possible.)

Ready to build!

Now I have a rough plan, and I’m ready to prototype!


Previous Post
How I plan to use AI to find a nanny share