https://wellfound.com/jobs/3365364-ai-engineer-remote-europe
Booksy is a cooperative library where the lent books are owned by the members, proud of its very curated collection of 20 books that can be lent by patrons. The collection is renewed each month. 5 books are kept, 15 books will leave the library, and 15 new ones will be added.
Using code-based, open-source tools, build a simple agentic workflow with two agents:
- a decision-making agent that will select the five books that Booksy should keep this month;
- a content personalization agent that will send an email to the owners of the books that won't be kept in the collection this month to explain the decision. We don't expect the emails to be sent, but they should be in your repository.
data.csv contains an example of what the agent's input data looks like.
Explain your design decisions and include instructions on how to run the code.
If you want to make calls to an external provider, you can use Anthropic, OpenAI or Gemini models.
If we have spare time, what other agents might make sense to add to this workflow?
How would you design a system to detect when the workflow starts making systematically bad decisions?
You're an AI Engineer at a SaaS company. Your customer support team has a distinctive, helpful brand voice, but the new AI chat assistant sounds "robotic and corporate." Customers complain that it doesn't match your human agents.
Describe a system to generate responses that match the style of each support member. Detail your architectural decisions.
Assuming your initial system works, but sometimes customers still complain responses are "cold and unhelpful." Describe how you would design a system that flags "corporate robot" responses before showing them to customers?