charity-donor-assistant

The Charity Donor Assistant is an AI agent that helps donors understand and compare charities. It is built on Amazon Bedrock with Anthropic Claude and uses tool calling to decide which action to take. The agent grounds every answer in real data with source citations and refuses to hallucinate—if a charity isn't in the database, it says so rather than inventing information. It was created as a hands-on project to explore production-style agentic patterns, RAG grounding, and trustworthy AI. The agent answers natural-language questions about charities while staying strictly grounded in verified data. It demonstrates agentic tool calling, RAG-style grounding, anti-hallucination, and reusable components. The agentic loop works as follows: the donor asks a question; the agent reads it and decides which tool to call; the tool retrieves real data from the charity database; the result is returned to the agent, which composes a grounded, cited answer. The agent has three tools: get_charity (look up one charity by name), compare_charities (compare two charities side by side), and search_charities (find charities by cause and/or minimum rating). The tools only ever return records from data/charities.json, each tagged with a source ID (e.g. CH001). The agent is instructed to cite those IDs and never rely on its own knowledge—so it physically cannot fabricate a rating.