Inspiration
Most automation tools assume you already know exactly what you want to build. But for many business owners, the biggest hurdle isn't the "how"—it’s the "what." We were inspired by the millions of pages of static SOPs, messy CSV logs, and complex policy documents that contain valuable business logic but remain "dead" on a hard drive. We wanted to build a bridge that turns passive documentation into active, executable software.
What it does
autoMate is an AI-driven "Discovery Engine" for business operations.
Discover: Users upload PDFs, CSVs, or URLs. Our RAG-powered agent analyzes these files to find hidden automation opportunities (like an "If-This-Then-That" rule buried in an HR manual).
Architect: Once an opportunity is selected, the system uses Context7 to fetch live node documentation and n8n MCP to verify technical patterns.
Execute: Powered by the Gemini 3 LLM, it generates a high-fidelity, "Plug-and-Play" n8n JSON workflow that users can immediately import and run.
How we built it
We built a multi-agent system using a robust, technical stack:
LangGraph: Orchestrates the "Super-Graph" that moves from document retrieval to code generation.
Gemini 3 LLM: Acts as the primary reasoning engine, utilizing its massive context window and advanced logic to architect complex workflows.
RAG (Retrieval-Augmented Generation): A custom pipeline that retrieves, grades, and transforms document context to ensure the AI doesn't hallucinate business rules.
Context7 & MCP: We integrated the Model Context Protocol to allow our agent to "talk" to n8n and fetch real-time documentation, ensuring our JSON parameters are 100% accurate.
Challenges we ran into
JSON Connectivity: Early versions of our agent generated nodes but failed to "draw the lines" between them because we didn't initially account for n8n's specific double-array connection syntax.
RAG Precision: Ensuring the AI didn't just summarize the text but actually extracted logic required building custom "graders" to filter out irrelevant information.
MCP Integration: Mapping a local AI agent to a cloud-based n8n instance without a Public API (for free-tier users) forced us to pivot from auto-deployment to generating portable, high-quality JSON artifacts.
Accomplishments that we're proud of
We successfully moved from unstructured text to structured code in under three minutes. Seeing a complex, multi-branching workflow appear in n8n—complete with specific data from a PDF the agent had just "read" via Gemini 3—was a massive win. We built a system that doesn't just respond to prompts; it offers insights.
What we learned
We learned that Context is King. An LLM can write code, but an LLM with access to specific documentation (via Context7) and a structured RAG pipeline writes reliable code. We also deepened our understanding of the Model Context Protocol, which we believe is the future of how AI agents will interact with third-party SaaS tools.
What's next for autoMate
Our next step is building the autoMate Studio—a modern, glassmorphic web dashboard that visualizes the "thinking process" of the agent in real-time. We also plan to add a "Self-Healing" feature where the agent can test the generated JSON against an n8n test instance and fix its own bugs before the user even sees the file.
Log in or sign up for Devpost to join the conversation.