Build end to end

Projects

A problem teaches one failure. A project takes you through a whole agent, step by step. Each step has a goal, a short explanation and a check that passes or fails on the same fixed model.

• Hard · about 8 hoursComing soonTrajectory grading engineA step-level evaluator that scores every tool call an agent makes against a deterministic DAG, and fails the run the moment it hallucinates an API parameter or skips a safety check.6 steps
• Hard · about 10 hoursComing soonProduction RAGRetrieval that holds up outside a demo: ingestion, hybrid search, reranking, citations, and an eval set that tells you when a change made answers worse.6 steps
• Hard · about 10 hoursComing soonPersonal agent with Jev at the centreA personal assistant over calendar, email and tasks where code owns the control flow, Jev decides at every fork, and LLMs only run when the output is words.7 steps