DocGen: Building an AI Assistant for System Analysts Without Losing Your Mind

A developer created DocGen, an AI assistant for system analysts, which processes stakeholder requests through a chain of seven LLM agents to produce a draft of software documentation. The system is split into two independent chains: the first handles context assessment and interview question generation, the second processes interview answers to produce a document. Between the chains, there is a human factor where analysts conduct interviews. The architecture includes Agent 0 for context quality assessment and temperature recommendation, Agent 1 for stakeholder-specific question generation, Agent 2 for normalization of raw answers into structured data, Agent 3 for draft generation, Agent 4 for validation and consistency checking, and Agent 5 for formatting. A clarification loop allows re-running the pipeline up to three times if validation finds issues. The developer added a hybrid mode in version 1.0.9, combining LLM text generation with programmatic PlantUML generators for architectural and UML diagrams to ensure syntax validity. The system supports eight document types including SRS, Architecture, API Spec, README, Test Plan, and UML diagrams. The author notes that the project took about a week of iterations and is currently in testing; they see AI as augmenting rather than replacing analysts, as verbal communication and human judgment remain important.