Fluent Pro — AI English learning
The brief
Fluent Pro had a vision and a learning-science team but no production engineering. They wanted a platform that gave students instant, personalized feedback on their English — pronunciation, grammar, fluency — at a fraction of the cost of a human tutor.
We built the platform end-to-end in 14 weeks. The core was a fine-tuned LLM eval pipeline feeding personalized feedback through a low-latency speech-to-text loop. Students upload a 30-second audio response; they get scored feedback in under 4 seconds.
The eval suite was the make-or-break work. Without rigorous evaluation, the LLM scored students inconsistently. We built a 600-item golden test set and a regression gate in CI so every prompt change had to clear it before deploying.
