AI-Native SaaS Boilerplate + Conversational Analytics Dashboard
A2SaaS — an open-source SaaS starter kit (npx create-a2saas) and a dashboard that answers plain-language questions about business metrics with live, inline visualizations
Overview
Two complementary products under one brand: a free, MIT-licensed SaaS boilerplate installable via npx create-a2saas, and a premium Insights Dashboard demonstrating conversational analytics. The boilerplate ships auth, payments, database, and email fully wired — plus an optional, deletable AI layer (chat, tool-calling, generative UI, and LLM observability) — structured so AI coding assistants can navigate and extend it without fighting abstractions. In the dashboard, users ask questions in plain language; an agent makes multi-step tool calls and renders each result as an interactive chart, table, or KPI card inline in the conversation.
Responsibilities
- Built a conversational-analytics experience with generative UI: Claude (via the Vercel AI SDK) makes multi-step tool calls, and each result renders as a live Recharts component — area/bar/waterfall/funnel charts, gauge KPIs, data tables — inline in the chat rather than as text
- Designed 14 callable data tools for querying metrics, MRR waterfalls, cohort retention, conversion funnels, channel comparison, and what-if scenario modeling
- Added LLM observability: distributed tracing (Langfuse + OpenTelemetry), thumbs-up/down feedback scoring, anonymous session tracking, and Anthropic prompt-caching to cut token cost and latency
- Designed an AI-friendly architecture — single schema file, flat structure, no hidden abstractions, end-to-end type safety — documented via a CLAUDE.md context file and executable skills so AI coding assistants extend it reliably
- Built a provider-agnostic adapter layer: swap Clerk → Auth.js, or the LLM provider (Anthropic → OpenAI), by changing one file
- Shipped a published npm CLI (create-a2saas) that scaffolds a zero-config project, with the AI layer as an opt-in, one-command-removable module
- Engineered a coherent 12-month mock dataset ($0 → ~$30K MRR) grounding every dashboard view and the AI's answers in realistic data
- Implemented demo guardrails: IP-based rate limiting and a lead-capture gate (Finance/Sales/Growth dashboards behind a demo-request form)
Primary Technologies
Outcome
Free, MIT-licensed, and live on npm via npx create-a2saas — now shipping an optional AI layer (chat, tools, generative UI, tracing) that works by adding a single API key or removes with one command. The Insights Dashboard runs as both a working product and a live reference implementation of the stack. Live at a2saas.com.
The site is live at a2saas.com, and a few screenshots of the app are on the landing page. The CLI is public on npm: create-a2saas. Contact me to see more.