A simple project with maximum optimization and cost-efficiency, built using available tools and free-tier services wherever possible.
- EC2
t3a.small, Reserved Instance for 3 years — $93 total (≈$2.58/month) - Additional: public IP ($3), EBS 25GB gp3 ($1.5), AWS Backup ($2)
- Total monthly cost — ~$6.5 (IP: $3, EBS: $1.5, Backup: $2)
- Can be made even cheaper:
- Remove AWS Backup
- Package everything in a Docker container or CloudFormation template for quick re-deployment
- Laravel 12
- Database: MySQL installed directly on the EC2 instance
- Backups:
- Daily dump using
mysqldump, gzipped to ~22MB, uploaded to S3 with 14-day lifecycle - Full AMI backups via AWS Backup
- S3 for backups (because why not — almost free)
- Daily dump using
- All static assets are stored within the Laravel project, with a backup on GitHub and locally. While best practice suggests storing assets in S3, it was decided to keep them local for simplicity.
- The entire site is cached
- Dynamic pages: cache expires every 2 hours
- Static content: longer expiration
- All optimized to reduce EC2 load
- Cursor download clicks are sent to Google Analytics as custom events via gTag
- Every 5 minutes, an API request pulls event stats
- Stats are updated in the database per cursor, adding to the previous count
- Installs/uninstalls are also tracked
- Upon uploading a cursor, several tasks run:
- Tag generation based on cursor and collection name
- Cursor and collection descriptions are generated
- Translated into 50+ languages via OpenRouter (Google models)
- A multilingual sitemap is regenerated monthly and submitted to Search Console
- Built using AWS Lambda + DynamoDB (fully within the Free Tier)
- Offloads even more traffic from the main server
- Initially used only CloudFront, but request limits led to added cost
- Final setup: Cloudflare in front of CloudFront
- Benefits: Caching, WAF, analytics, and request offloading
- Cloudflare handles both edge traffic and static content, with CloudFront acting as an additional layer
- The pipeline is intentionally simple and efficient.
- It triggers an AWS Lambda function, which then uses SSM Run Command to pull the latest changes directly on the production EC2 instance.
- Deployment happens almost instantly.
- For this project, it’s the most straightforward and effective pipeline possible — minimal overhead, near-instant execution, and fully server-driven.
- Meilisearch is containerized and separated from the main app
- One index per language
- Cursors are added to Meilisearch shortly after upload, as part of the update flow
- Metrics collected via Prometheus
- Displayed in Grafana
- Plan to eventually render all metrics directly in the Laravel admin panel (which already includes a lot of analytics)
- 3 containers: chat server, MongoDB, and toxicity filter
- Toxicity detection:
- Custom ML model trained on 8M comments (very fast)
- If passed, then checked again via OpenRouter (Haidy AI)
- Efficient across multiple languages
- Port 22 on EC2 is closed
- SSH access via WARP tunnel (Cloudflare Zero Trust)
- Acts like a VPN, but more secure and modern
This project is for educational purposes only. Do not use any resources or code without explicit permission.