This dataset includes traces and associated metadata from agentic interactions on a Financial Reasoning task. We built the agentic system in langgraph and ReAct agents. In each sample, the agent is given a question that must be answered with information from one or more company 10-K's. The 10-K's are accessible to the agents through tools.
Our question-answer pairs have been carefully co-created with Snorkel's Expert Data-as-a-Service network of financial… See the full description on the dataset page:
https://huggingface.co/datasets/snorkelai/agent-finance-reasoning.
