Poppins used Basalt to empower product teams to handle prompting, analyse logs, and iterate 2x faster on their product. Using the prompt prototyping and AI powered evals, they were able to go from draft to production in a few days.
Evaluations
Test in workbench
Prompts in prod
Prompt versions
Poppins used Basalt to empower product teams to handle prompting, analyse logs, and iterate 2x faster on their product. Using the prompt prototyping and AI powered evals, they were able to go from draft to production in a few days.
LLM features are becoming core in their product, allowing users to seamlessly create a new listing and empowering content moderators to root out frauds. With Basalt, they iterate on prompt quality, monitor performance, and test models to optimize both outcomes and efficiency.
Before adopting Basalt, the team struggled with uncertainty in their prompt engineering process
"Some prompts generated good results, some didn't... but we had no idea why, or how to systematically improve them," explains Loick, cofounder and CTO.
This lack of visibility made it challenging to maintain consistent quality across their AI features. Now, armed with detailed analytics and testing capabilities, they iterate continuously with concrete, data-driven feedback. A prime example of their improved process can be seen in their content moderation feature. After carefully optimizing the prompt structure and parameters with Basalt's testing framework, they've achieved remarkable results: zero customer complaints about moderation decisions. Nada. 🎉
With Basalt, they shifted AI ownership towards product teams and prompts are directly managed by PMs.
AI in our app is made of product rules. So it should be owned by product
☑️ Test multiple models per prompt to reduce both $ cost and carbon footprint.
☑️ Evaluate prompt quality even outside the app (e.g. in tools like Cursor).
☑️ Integrate Basalt into more product workflows, continuously.
Evaluations
Extensive use of LLM-as-a-judge evaluators
Test in workbench
Multiple batch runs accros hundreds of test cases
Prompts in prod
Multiple AI prompts in production in less than a month
Prompt versions
Heavy iteration on prompts to reach high quality standards
The platform to integrate AI in your product in secondsGo from idea to ready-to-deploy prompts