What I Learned in 7 Days
the whole project in one picture, a clean-up checklist to avoid surprise bills, and what I learned.
We made it
Seven days ago I'd never used Amazon Bedrock. Today I have a chatbot that answers from my class notes, shows its sources, and has safety checks.
Last step: tidy up, so nothing keeps costing money.
What I built
My notes in S3 → Knowledge Base finds the right chunks → a Bedrock model writes the answer → Streamlit shows it with sources. Guardrails check the question and the answer.
This is called RAG: find first, then answer. It can improve relevance, but it doesn't guarantee every answer is correct.
Clean up
Stopping Streamlit only stops the app on your laptop. The AWS resources behind it are still there.
If you no longer need them, review each one, in this order:
- Stop the app (Ctrl + C).
- Delete the Knowledge Base.
- Delete its vector store (for example, the S3 Vectors bucket). Deleting the Knowledge Base doesn't always delete this.
- Delete the guardrail.
- Empty, then delete the notes bucket, if it's only for this project.
- Remove IAM keys or users made just for this project. Never delete shared or default resources.
- Check Billing after a day or two, and keep your budget alert on. Alerts warn you, but they don't stop spending.
What can cost money
Model calls, embeddings, the vector store (some options have ongoing charges), guardrails, S3 storage, and logs if you turned them on.
Check your actual usage. Don't assume anything is free. See Bedrock pricing and S3 pricing .
5 things I learned
- Prompts matter, but they don't guarantee correct answers.
- Embeddings turn meaning into numbers, so AI can find similar ideas.
- RAG connects answers to your own notes.
- A chat screen is only one part of an app. Permissions, retrieval, and errors matter too.
- Safety needs testing. Guardrails reduce risk, but they aren't perfect.
Conclusion
In seven days I went from "What is Bedrock?" to a working AI chatbot. The biggest lesson: an AI app is more than a model call. It needs data, retrieval, an interface, safety checks, testing, and clean-up.
I'm still learning, but I built this myself. If you've been waiting to start, pick one small project and build it a day at a time.
Thank you for following 7 Days of Amazon Bedrock. I hope its helpful for you
check my series
Day 1 : Meet Amazon Bedrock
Day 2 : 5 Ways to Get Better AI Answers
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article