viggnah/wso2-africa-hackathon-kit

# WSO2 Africa Hackathon — Developer Starter Kit Build an **Enterprise AI Assistant powered by WSO2** that solves a real problem for an industry you choose. It should securely connect to enterprise systems, invoke APIs, automate workflows, and provide intelligent, actionable experiences. Pick any industry and compose the products however you like — **there is no prescribed architecture.** This kit gets you productive fast, even if you've never touched a WSO2 product, and is designed to work *with* AI coding tools (Claude, Cursor, Codex, etc.). Read this page once, install the skill into your AI tool, and start building. ## What's in this kit | File | For | Use it to | |---|---|---| | **This README** | You | The map and human-readable overview: prerequisites, product tour, how the pieces fit, composition patterns, first-hour setup, and what judges reward. **Start here.** | | **Skill** — `.claude/skills/wso2-ai-assistant/` | Your AI tool | Install as a skill/rules file so the model loads the correct WSO2 mental model and fetches docs instead of inventing APIs. `SKILL.md` is always-on; `references/` holds the deep detail (products, composition patterns, setup) the model loads on demand. **Install per tool** — see below. | ## Before the event You don't need to be an ML researcher, but arrive comfortable with the shape of agentic AI and with a working machine. ### Concepts worth knowing If any of these are new, spend an hour on them before day one. Skimming the docs/quick-start of whatever agent framework you plan to use helps too. - **LLM basics** — what a prompt / token / context window is. - **Tool use / function calling** — an LLM deciding to call an external API and using the result. - **The agent loop** — plan → act → observe → repeat. - **Memory** — conversation history vs. persisted state. - **RAG** — retrieving real data to ground answers instead of hallucinating. - **MCP** — a standard way to expose tools and data to an agent. - **Guardrails** — PII mas …