Human-AI Hybrid Mathematical Discovery Workflow: Unifying Ramanujan's 17 Series for $\boldsymbol{\frac{1}{\pi}}$

We present a human-AI hybrid research workflow and apply it to a long-standing challenge: unifying Ramanujan's 17 remarkable series for $\frac{1}{\pi}$ (1914) under a single mathematical structure. The workflow combines human steering, a persistent LLM-based agent equipped with project-specific context, symbolic and numeric computational guardrails, and custom *skill.md* files. Special emphasis was placed on verification efficiency. Automated independent test patterns were employed, and generated artifacts were optimized for rapid human interpretation and evaluation. Both successful strategies and characteristic AI failure modes are discussed. This methodology yielded a new, elementary proof that all 17 formulas arise from a single algebraic template: a Conservative Matrix Field (CMF), where varying just a few parameters recovers each of Ramanujan's formulas. The agent and skill files have been open-sourced.