A large technology client needed domain-specific demonstrations to improve LLM performance using reinforcement learning techniques. The work required Finance (US) expertise to ensure the demonstrations reflected accurate financial context. The demonstrations also needed to be produced in English and aligned to reinforcement learning workflows. Finance (US) experts were engaged to produce English-language demonstration data tailored for reinforcement learning use. The demonstrations were created to fit the client’s RL data requirements and support model performance improvements. The delivery focused on producing a consistent set of demonstrations suitable for RL workflows. A total of 3,500 datapoints of Finance (US) demonstration data were delivered for the project. The dataset was produced in English and aligned to the client’s reinforcement learning workflow needs. This provided the domain-specific demonstrations the client required for its reinforcement learning data pipeline.

A big tech client needed high-quality multilingual demonstrations to support RAG-focused post-training. The customer required consistent, well-edited data suitable for post-training foundational LLMs. The scope included multiple languages, increasing complexity and quality requirements. Skilled editors created multilingual demonstration datasets for the customer. The datasets were produced in English, German, and Italian to support the RAG-focused post-training work. The delivered content was prepared for use in post-training foundational LLMs. The project delivered demonstration datasets across three languages. A total of 2,500 datapoints per language were delivered for the post-training effort. The customer received multilingual demonstrations aligned to RAG-focused post-training needs.