Developer ditches LLM for 200-line template engine to build letter generator

A developer building a letter-generation feature chose deterministic template code over an LLM API call, citing cost, reliability, and document sensitivity as key reasons. The engine handles resignation, notice, and recommendation letters across formal, warm, and brief tones using conditional paragraph assembly and simple switch logic. Because output shape and content are fixed, the solution is fully testable, works offline, and carries zero per-use cost. The approach also enables strict input validation that differs by letter type, catching missing fields before any output is generated. The developer acknowledges the method cannot handle open-ended or variable prose, but argues short formal documents are better served by deterministic code than generative models.