Cedar Valley Spring Count Intake

In this project, you'll build a tool-using LLM agent that turns messy, multi-source records into one clean, validated dataset plus a review queue. The data is a volunteer bird count whose notes arrive in three forms: a paper logbook, an email, and a legacy spreadsheet. The model reads the mess; the tools you build enforce the rules. Along the way, you'll define a Pydantic model that doubles as a validator, build a class to hold the run's state, and expose four tools through an in-process MCP server. You'll write an agentic loop that commits each record through an authoritative gate and sends anything ambiguous to human review instead of guessing. This ties together function calling, MCP, structured output, and object-oriented design into the pattern behind real LLM systems, and makes a strong portfolio piece for any AI engineering role.