Summary and Closing Remarks

Between the previous video and this one, the project was polished further. Let's look at what changed and the final result.

Post-recording improvements

Several improvements were made after recording:

• PostgreSQL logging was finished (there were timestamp issues with Docker clocks being out of sync with the host)
• Grafana dashboard provisioning was automated with an init script that creates the data source and loads the dashboard JSON
• The README was polished with a generated banner image, better formatting, grammar fixes, and clearer instructions
• Code readability improvements across all files

Total cost

The entire project cost about $2 in OpenAI API calls:

• Dataset generation: ~$0.50
• Ground truth generation: ~$0.50
• RAG evaluation: ~$0.50
• Testing and debugging: ~$0.50

Using GPT-4o-mini keeps costs low. You could reduce costs further by using a local model or a cheaper provider.

Tips for your project

Keep these points in mind as you build your own project:

• Start simple: get a basic RAG flow working first, then add evaluation, monitoring, and containerization
• Use generated data if you don't have real data - it's good enough for a course project
• Polish your README: it's the first thing people see
• Use the course project criteria as a checklist

The final project

Check out the completed project at alexeygrigorev/fitness-assistant.

It includes:

• A fitness exercise dataset generated with GPT
• Search with minsearch
• RAG flow with OpenAI
• Retrieval and RAG evaluation
• Flask API
• Docker Compose with PostgreSQL and Grafana
• Automated Grafana provisioning
• A polished README

← Monitoring and Containerization | Chunking for Longer Texts →