Lovable Clone — AI Agents for App Generation (LangGraph)
A simplified clone of Lovable.dev where an AI agent generates, builds, previews, and fixes a working app from a natural-language spec.
Workflow & Architecture
How this system works
User describes the app in natural language via the control panel.
LangGraph agents synthesize requirements and scaffold a Next.js project.
Code generation, shell tools, and error parsing in an iterative fix loop.
Working app rendered in an in-browser iframe for immediate review.
Overview
Designed a LangGraph state machine (spec synthesis → planning → scaffolding → build → fix → preview) driving a Next.js scaffolding template, with a FastAPI + HTMX control panel for chat, task logs, and a live iframe preview.
The Problem
App generation workflows need multi-step reasoning, planning, code generation, build checking, and iterative fixes from a plain-language idea.
The Solution
Built a LangGraph state machine with an edit loop for follow-up prompts, supporting tool agents for file handling, shell commands, build-error parsing, and project export.
Key Features
- Multi-step LangGraph agent pipeline from spec to working app
- Edit loop for follow-up prompts with automatic re-scaffolding
- Supporting tool agents: file handling, shell/server commands, build-error parsing, project export
- Live in-browser preview of the generated app
- FastAPI + HTMX control panel with task logs
Outcomes & Business Value
- End-to-end agentic pipeline from natural language to deployable artifact
- Automated build/fix loop for iterative app generation
- Demonstrates multi-step agentic software creation workflow
