This mindmap is an output from Genie, Emumba's dedicated GenAI R&D group. It is a practitioner's map of the GenAI application development landscape, organized around the software development lifecycle. Rather than cataloging use cases (chatbots, copilots, search, etc.), it decomposes the space into the underlying capabilities, decisions, and tooling that compose any GenAI system — so teams can navigate to what's relevant for what they're building. The three top-level stages reflect how a team actually works: Design & Development (what you build and how), Evaluation (how you know it works), and Operations (how you run it in production). Use it as a decision map when architecting a new system, an audit tool when reviewing an existing one, or a shared vocabulary across engineering, product, and research.
Design & Development. GenAI applications vary enormously in form — a RAG-powered search tool, a voice agent, a document classifier — but they are all composed from a common set of building blocks. Rather than organizing by use case (which would leave most of the map irrelevant to any one team), this section is organized by capability: the distinct, independently designable parts of any GenAI system. Most applications combine several of these — a coding assistant might use Model Invocation, RAG, Tool Execution, and Agents together — while simpler apps may need only one or two. Navigate to the capabilities your system uses; the decisions and tradeoffs within each apply regardless of your specific product.
Evaluation. Evaluation is organized to mirror how well-functioning teams actually test GenAI systems: first by what you're trying to measure (strategy, accuracy, performance, safety), then within accuracy, by which capability you're evaluating. This matters because retrieval quality, tool-call correctness, agent trajectory, and model output all fail in different ways and require different metrics and data. A single "does it work?" eval misses most real failure modes. The recommended approach is capability-scoped testing at each stage of development — not just end-to-end testing before launch — treating eval as an ongoing engineering discipline rather than a pre-deployment checklist.
Operations. Running GenAI in production: where models and tools run, how you observe them, and how you control cost and capacity. Overlaps with classic DevOps but adds LLM-specific concerns (tokens, embeddings, agent traces).
Click nodes to expand / collapse · Scroll to zoom · Drag to pan