Rapid improvement in GenAI models using high quality, multilingual STEM content, with 100% quality compliance for a global tech giant

Introduction:

A leading global provider of search services, entertainment platforms, and cloud computing services partnered with Firstsource to create education-focused content to train its GenAI models by leveraging Reinforcement Learning from Human Feedback (RLHF), Human Factor Validation, Side x Side Evaluation, Human-in-the-Loop process and factuality checks. The focus areas included STEM content creation, localization, creative writing and content validation.

Challenges:

The client sought to train their GenAI model with multilingual STEM content and general knowledge content for a global user base, including students while ensuring a quick turnaround, strict compliance standards and consistent quality.

How we made it happen:

• Content Creation: Developed STEM content across various difficulty levels for students from grade 4 to university level, covering subjects like English, Physics, Chemistry, Math, Biology, Medicine, Computer Science, Business & Government.
• Localization into multiple languages: Created and translated educational content into Hindi, Indonesian, Chinese, Japanese, French Spanish, Portuguese, Italian, Turkish & German languages to cater to diverse user groups.
• Onboarding experts with varying skillsets— Content Creation, Review, Rewriting and validation done by Undergrads, Graduates, Masters, Comparable Experience & PhDs.