You don't think about your Streamlit app optimization until you deploy it to AWS
summary
A Streamlit kana OCR prototype using the Hugging Face model "mango" loads repeatedly per session because Streamlit session_state doesn’t survive new tabs, so the fix is preload the HF model during image build to avoid repeated downloads and per‑GB NAT Gateway charges (~$0.05/GB). Deployment uses ECS Fargate tasks across two AZs with autoscaling, CloudFront + ACM TLS + Shield + WAF for DDoS and WAF rules, and use AWS Secrets Manager and Cognito for credentials and auth, which proves reliable.
AI-written summary · about
chapters (8)
- 0:00Streamlit app optimization
- 3:19What is Streamlit
- 5:07Model loading performance
- 11:57Autoscaling challenges
- 17:29Security for Streamlit apps
- 21:23Credential storage with AWS
- 23:56Key takeaways
- 25:22Q&A
related talks
39mAdvanced Streamlit for Python Developers
40mCreating Interactive GenAI applications powered by Streamlit, open source LLMs
27mBuilding Serverless applications with Python
30mDeploying scalable Django application on AWS
50m Serverless Python in the real world: common patterns & challenges 
41mApplying serverless architecture pattern to distributed data processing
07mBackstage interview
26mStreamlit - Build interactive data dashboards quickly