Domain Specific Supervised Fine-Tuning
Teaching the teacher! Our clients wanted to train an LLM to act as a personalized tutor who would teach, engage and encourage students. The model had to be fine tuned on 25+ domains from various fields such as humanities, business, engineering, and sciences. This was a time sensitive ask to stand up 500+ world-class domain experts within 2 weeks.
We followed a two-pronged approach:
1. Train the model on tutor tone/style and conversational skills (while also helping it with technical skills): This was accomplished via creating large dataset of multi–turn conversations .
2. Further fine-tune the model to have very high domain accuracy: Create narrower, even higher-quality data in the 25+ domains. This dataset was built with a heavier concentration of PhD domain experts
We successfully identified, trained, and assigned the initial set of 500+ experts within 2 weeks (delivering 1000+ hours of fine tuning per day). All with very high acceptance rates and very low fraud rates.
Given strong initial success, we were quickly pulled in to help with the entire project. The tutor model is being prepped for market go-live in this quarter (Q2 ‘24)
Reach out to us at [email protected] for more information, work samples, etc.