Final Assessment

1.API fundamentals — I can make authenticated requests, manage multi-turn conversation history, and use system prompts, temperature, streaming, and structured output extraction via prefill + stop sequences.

2.Prompt evaluation — I can build an end-to-end eval pipeline with generated datasets, model graders, and code graders, and I know how to iterate prompts against the resulting score.

3.Prompt engineering — I can apply clarity and directness, specificity with guidelines and process steps, XML structure, and one-shot or multi-shot examples, measuring each technique's lift with evals.

4.Tool use — I can design tools, write JSON schemas, handle multi-block messages, build conversation loops with error handling, and integrate built-in tools like the text editor and web search.

5.RAG and retrieval — I can chunk documents, generate embeddings with VoyageAI, store them in a vector index, and build a hybrid retriever combining vector search and BM25 with reciprocal rank fusion.

6.Advanced features — I can use extended thinking, images, PDFs, citations, prompt caching with breakpoints, the Files API, and the code execution tool as production capabilities.

7.Model Context Protocol — I can build MCP servers with FastMCP, expose tools, resources, and prompts, and wrap ClientSession in a custom MCPClient that handles discovery and invocation.

8.Claude Code — I can install, configure, and use Claude Code effectively with /init, CLAUDE.md memory files, the context-plan-implement workflow, and MCP server enhancements.

9.Agents and workflows — I can apply the four workflow patterns (evaluator-optimizer, parallelisation, chaining, routing), design abstract combinable tools for agents, and choose between workflows and agents based on task definition.

10.Overall, I feel confident designing and building a production Claude application that combines several of these capabilities.