Projects
Production-grade agent systems built incrementally through the book.
Ingest. Retrieve. Cite. Escalate on uncertainty.
Side-by-side comparison of raw, ADK, and LangChain agent implementations on identical queries to quantify framework overhead.
Inspect signals, search runbooks, propose remediation, request human approval.
Hands-on experiments that make token counting, cost projection, and structured output tangible before you build agents.
Memory-augmented pipeline with session, long-term, and shared memory layers.
An instrumented multi-step agent loop with per-step cost logging, exportable JSON traces, and graceful error recovery.
A single-turn assistant that selects and executes tools with Pydantic validation, isolating tool-call logic from the agent loop.