Dev-Tools SaaS and Agentic AI Studio
Six months of sole-engineer ownership across a visual blueprint-and-logic-flow platform with an embedded AI studio. Database, security, the agentic studio, a visual type-system editor, realtime sync, testing, CLI, and release automation.
The challenge
An enterprise dev-tools company needed one engineer who could own the full architecture: let non-developers author production business logic visually with AI assistance that can directly read and mutate the model, on a secure multi-tenant foundation, closing the loop from natural-language requirement to running, tested software.
The solution
Took full architectural ownership over six months (79 commits, roughly 323k insertions across the database, security, AI features, visual editors, realtime, CLI, and CI). Migrated the data layer to Supabase/Postgres, then hardened multi-tenancy so row-level security is enforced on every operation. Built a genuinely agentic AI studio: a tool registry of model-mutation tools exposed through both an MCP adapter and the Vercel AI SDK, driving a bounded multi-step tool-calling loop, fronted by an LLM security check that screens input for prompt-injection before any generation runs. Added a visual type-system editor on React Flow, realtime blueprint sync, a Storybook suite, Playwright E2E, a CLI with OAuth PKCE, and an automated release pipeline.
Results
- Six months, 79 commits, one engineer across every layer
- Data layer migrated to Supabase/Postgres; multi-tenant workspaces with row-level security enforced platform-wide
- Agentic AI studio with a model-mutation tool registry over MCP and the AI SDK, plus a prompt-injection security gate
- Visual type-system editor on React Flow with realtime blueprint sync
- About 428 fast Storybook component tests and a deterministic Playwright E2E suite, from zero
- CLI with OAuth PKCE and an automated, LLM-assisted release and changelog pipeline
Want similar results?
Let's talk about your project and how I can help.