-
-
Landing Page
-
Cross reference of each study session
Inspiration
When my friends and I are in different study groups, we usually want to compare notes, since a lot of the time random, super helpful info pops up or someone has an interesting perspective.
What it does
With Bee, we compare the two study group convos and show what was said differently, which topics weren't covered, and any important action items that came up.
How we built it
I used Claude Code to build it with:
- Bee CLI: each person scans a QR code to log in and picks one convo
- Claude API: Haiku 5.5 turns each transcript into notes, Sonnet 5.5 compares them
- AWS: CloudFront + S3 for the site, API Gateway + Lambda running the Bee CLI, a worker Lambda for the AI part, SSM for secrets, all set up with Terraform
- Podman/Docker Compose for running it locally with just an API key
Challenges we ran into
- Bedrock was blocked on our account ("Access to Bedrock models is not allowed"), so we switched to the Claude API with Bedrock as a fallback
- Bee login in a serverless setup: the login flow expects to keep running, so we encrypt each person's session and keep it only in their browser tab
- The AI kept "fact checking" our notes with its own knowledge, so now every point has to point back to an actual line from the convo
- First AWS deploy hiccups: a Lambda concurrency limit of 10 and S3 permissions
Accomplishments that we're proud of
- Nothing is saved on our end: no accounts, no stored logins, transcripts deleted right after use
- It only shows the differences, so it's quick to skim
- Getting this up and running in a super short time
What we learned
- Learned a bit more about things to watch out for when using Lambda
- Deploying AWS services using Terraform
What's next for Cross Pollinator
- More than two people/groups at once
- Push action items straight into everyone's Bee todos
- Move to Bedrock once it's enabled on our account
- Export to flashcards (Anki) or Notion
Built With
- node.js
- terraform
- typescript
