DEV Community

naoanao
naoanao

Posted on

I Built an AI Clone of Myself to Run My Restaurant's Marketing While I Sleep (and Sold the Blueprint for $49)

#ai #python #indiehacker #webdev

I'm a restaurant owner in Japan. For most of 2025, I was spending 3+ hours every week on competitor research — manually checking what nearby restaurants were posting, what events were coming up locally, what food trends were emerging.

I couldn't keep up with social media, blogging, AND running a physical restaurant.

So I built two things.

What I Built

1. Growl — AI Marketing Copilot for Restaurant Owners

Growl connects your restaurant's menu, reviews, and local market data to generate:

  • Weekly 3-action marketing plan — the highest-ROI moves for YOUR restaurant this week
  • 7 days of social content ready to copy-paste
  • Review responses that match your restaurant's tone
  • Competitor intelligence — what other local restaurants are doing

The 3C competitor analysis that used to take me 3+ hours? It now takes 2 minutes.

2. Sage Blueprint — My Autonomous AI Content System ($49 one-time)

After building Growl, I realized I was still manually posting to social media, writing blog articles, and running market research for my own brand.

So I built Sage — an AI clone of myself that runs 24/7:

  • Posts to Bluesky/social daily, in my voice
  • Publishes a blog article every morning at 9am
  • Scans the market for trends at 3am
  • Generates content ideas based on past performance

The "identity layer" is three files:

  • identity.json — your niche, tone, audience, brand name
  • SOUL.md — your values, communication style
  • HEARTBEAT.md — your autonomous schedule

Change these files → everything shifts. It's not a tool you manage. It's a clone of you.

Running in production since January 2026. Sage has now published 360+ blog articles, 700+ social posts, and conducted daily market scans — all without me touching it.

The Tech Stack 

Python/Flask → LangGraph → Groq (Llama 3.3 70B) → Cloudflare Workers
Enter fullscreen mode Exit fullscreen mode

Key design decisions:

Why Groq instead of OpenAI? Speed and cost. Groq runs Llama 3.3 70B at ~500 tokens/second. For a system that needs to generate content 10+ times per day, latency matters.

Why LangGraph? The autonomous agent needs to make decisions — "should I post today or skip?", "is this content good enough?". LangGraph's conditional routing handles this cleanly.

Why Cloudflare Workers? My PC doesn't stay on 24/7. Cloudflare Workers act as a persistent scheduler that pings my local Flask server via ngrok tunnel when it's running, and handles the schedule when it's not.

The "human-like" layer: The system randomly skips 20% of posts, adds 2-40 minute random delays, and takes 2 random "days off" per week. This prevents the mechanical regularity that gets flagged by algorithms.

Results (Honest Numbers)

  • Growl: Just launched on Product Hunt (May 2026). 3 upvotes. Cold start is real.
  • Sage Blueprint: Listed on Gumroad at $49. Link here
  • Content output: 360+ articles, 700+ social posts generated autonomously
  • Time saved: ~3 hours/week on competitor research alone

Revenue so far: $0. Working on distribution now.

What I Learned

Building the system was the easy part. Distribution is where I'm stuck. The product works. Getting it in front of the right people is the actual problem.

If you're building something similar or you run a restaurant/small business and want to try Growl free, I'd love honest feedback.

Happy to answer technical questions about the LangGraph architecture, the Groq integration, or how the autonomous scheduling works.

I'm documenting this entire journey publicly. If you want to follow along:

Top comments (0)