How We’re Building Shy Car: AI Automation, Prompt Chaining & Indie Hacker Tools

A behind-the-scenes look at how we’re using APIs, AI agents, and webhooks to help humans (awkwardly) sell their cars — and what it means for building in the age of robots.

🧵 Introduction

In the Age of Robots, a lot of people are focused on building large, complex systems. But sometimes, the magic is in building something very small — a tool that does one thing well, over and over, with the help of automation and AI.

That’s what we’re doing with Shy Car — a project that helps people sell their used cars without sleaze, dealerships, or awkward classified ads.

At first glance, it looks like a car-selling tool. But under the hood, it’s a real-time, AI-powered publishing system — dynamically generating car listings tailored for Craigslist, Facebook, and private sale groups using a combination of:

VIN decoders
webhook triggers
prompt chaining
dynamic persona shaping
and some spicy middleware glue

Here’s how we built it.

🏗️ Step 1: The Core Idea – Human + AI Matchmaking

Most people don’t know how to write a good car listing.

We do. Or, more accurately — the AI does.

The idea behind Shy Car is to create a “dating profile for your car”, based on real-world inputs:

VIN (for base data)
Images (optional, for human review)
Seller goals (speed, price, low hassle)
Marketplace targets (Craigslist, Facebook, OfferUp, etc.)

🔌 Step 2: VIN Decoding via API

We use a third-party VIN Decoder API (currently NHTSA for base proof of concept, or commercial providers like VinAudit) to gather:

Make, model, year
Body type, trim, fuel type
Engine spec
Manufacturer info

Once a user inputs a VIN, we call the decoder API via GET, and parse the response to a normalized internal vehicle schema, so our AI agent can handle edge cases (like partial data, missing trims, etc.).

Example VIN call:

GET https://vpic.nhtsa.dot.gov/api/vehicles/decodevinvaluesextended/<VIN>?format=json

We use this to populate a simple object like:

{
  "year": 2013,
  "make": "Toyota",
  "model": "Highlander",
  "trim": "XLE",
  "fuelType": "Gasoline",
  "driveType": "AWD"
}

🔁 Step 3: Webhook-Triggered AI Flow

Once VIN data is collected, we trigger a webhook that kicks off the AI writing process.

Our stack:

Frontend: JavaScript widget or form inside Brilliant Directories (or standalone React component)
Middleware: Pipedream / UChat / Pabbly / custom Flask endpoint
AI model: GPT-4o-mini (for fast + cost-efficient generation)

We send a structured payload to the AI agent:

{
  "vehicle": {
    "year": 2013,
    "make": "Toyota",
    "model": "Highlander",
    "trim": "XLE",
    "mileage": "123,000",
    "driveType": "AWD"
  },
  "platform": "Facebook Marketplace",
  "voice": "quirky, honest, lightly awkward"
}

This is routed via webhook to the OpenAI API, with a custom prompt designed like so:

🧠 Prompt Engineering Example:

You are a helpful assistant writing car listings for private sellers.

Write a listing for this vehicle: {{vehicle.make}} {{vehicle.model}} ({{vehicle.year}}).

Platform: {{platform}}

The tone should be honest, human, and a little quirky. Mention highlights (AWD, mileage, etc.) and gently suggest why this car is a great next chapter for someone.

Include a title, description, and 3 hashtags. Do not add contact info.

The result is a context-aware, voice-driven listing, optimized for the platform. We tweak the prompt slightly based on the target channel:

Craigslist = more structured + serious
Facebook = casual + emoji-friendly
Dealers = formal + feature-forward

📬 Step 4: Delivery & Post Options

Once the listing is generated, the system either:

Displays it directly in the browser (preview + copy to clipboard)
Emails it to the user
Saves it to their account in Brilliant Directories
Sends it to a Pabbly flow to auto-post (optional)

🧱 Bonus Tools in the Stack

Paperform: Front-end lead intake / embedded form styling
UChat: AI-based user interactions (coming soon)
Pabbly Connect: API orchestration, connecting Brilliant Directories → AI → user response
Cloudflare Pages: For hosting edge functions + experimental features

🤖 Why This Matters in the Age of Robots

This is the new indie builder stack.

Instead of:

Writing code from scratch
Building databases
Designing custom UIs

We’re building with APIs, prompts, and agents.

The “product” isn’t just a car listing generator. It’s a tiny AI toolchain:

Input → Enrichment → Generation → Delivery
That’s the new loop.

If you’re a builder, this is the model:

Think smaller.
Build smarter.
Let the robots do the boring parts.

🧪 What’s Next?

We’re expanding Shy Car with:

Photo analysis + alt text generation
Instant dealership outreach packages (PDFs)
VIN-based car ranking tools (e.g. “best years to sell”)

All powered by chained AI actions + pre-trained prompt modules.

We’ll document that as it happens — right here at Age of Robots.

🎉 Final Thoughts

If you’re a builder, remember this:

In the age of robots, it’s not about doing more — it’s about doing the right thing faster, using the tools we have.

Shy Car isn’t a startup. It’s an experiment in automated empathy.

And we’re just getting started.

Want to build something similar?

Drop your questions in the comments, or subscribe for more build logs.

Ideas Bots

How We’re Building Shy Car: Automation, AI, and the Future of Selling Stuff