What if you could see your startup's failure before it happened?
Most startups don't die from bad luck. They die from the same silent killers — wrong market, wrong timing, wrong assumptions — that nobody caught early enough. Startup Autopsy gives founders a mirror to look into the future.
Video Link: [https://drive.google.com/file/d/1HCjj941-zbRX7_gjPIvyRWXkDKIga3aY/view?usp=sharing]
Every year, thousands of founders pour months (and millions) into ideas that fail for predictable reasons.
The painful part? Most of those reasons were visible from day one — in the pitch, in the assumptions, in the market signals. But there's no tool that stress-tests your idea before you build it.
Advisors are expensive. Feedback is biased. And post-mortems only happen after the damage is done.
Startup Autopsy is a multimodal AI agent that reads your startup like an investor, thinks like a skeptic, and writes like a documentary filmmaker.
You feed it your idea — a description, a pitch deck, screenshots, whatever you have — and it generates a future post-mortem: a plausible, detailed simulation of how and why your startup could fail, and what early signals you should be testing right now.
Think of it as a stress test for your conviction.
-
Input your startup
Paste your idea description, upload your pitch deck, or drop in screenshots of your product or landing page.
-
The agent analyzes
It processes your inputs across multiple dimensions:
- Market assumptions
- Competitive landscape
- Team blind spots
- Go-to-market risks
- Timing
-
Get your autopsy
A narrative post-mortem simulation that walks through the plausible failure arc of your startup, written in the style of a documentary.
Not to scare you — to sharpen you.
-
Extract early signals
Every autopsy ends with the specific assumptions you must validate before going further.
Most feedback tools give you:
- a score
- a checklist
- generic “risks to consider”
That’s not how humans internalize truth.
Stories do.
When you read a post-mortem written about your startup — even a simulated one — it hits differently. It forces you to confront the version of events you’ve been avoiding.
The insight behind Startup Autopsy:
Narrative is a stronger forcing function than analysis.
- Frontend — React + TypeScript (Vite)
- AI Agent — Multimodal Claude (text + vision)
- AI Platform — Vertex AI (Google Cloud)
- Deployment — Docker-ready
# Clone the repo
git clone https://github.com/anshi312/startup-autopsy.git
cd startup-autopsy
# Install dependencies
npm install
# Start the dev server
npm run dev
## Vertex AI Authentication (Docker Setup)
This project uses **Google Vertex AI with Application Default Credentials (ADC)**.
Instead of API keys, the container uses your local Google Cloud authentication.
---
### 1. Host Setup (One-time)
Run these commands **on your computer (not inside Docker)** to generate the credentials file.
| Action | Windows (PowerShell) | Mac (Terminal) |
|------|------|------|
| Login | `gcloud init` | `gcloud init` |
| Generate Credentials | `gcloud auth application-default login` | `gcloud auth application-default login` |
| Verify File | `Test-Path "${env:APPDATA}\gcloud\application_default_credentials.json"` | `ls ~/.config/gcloud/application_default_credentials.json` |
This creates the **Application Default Credentials file** used by Vertex AI.
---
### 2. Run Docker Container
Mount the Google Cloud credentials directory into the container.
#### Windows (PowerShell)
```powershell
docker run -it `
-p 8080:8080 `
-v "${env:APPDATA}\gcloud:/root/.config/gcloud" `
-e GOOGLE_APPLICATION_CREDENTIALS=/root/.config/gcloud/application_default_credentials.json `
your-image-name
#### Mac / Linux / Unix based system
``` docker run -it \
-p 8080:8080 \
-v "$HOME/.config/gcloud:/root/.config/gcloud" \
-e GOOGLE_APPLICATION_CREDENTIALS=/root/.config/gcloud/application_default_credentials.json \
your-image-name
### Key Rules
---
- **No API Key** — Do not use `process.env.GEMINI_API_KEY`. The SDK automatically detects the mounted credentials.
- **Use Vertex AI** — Ensure your AI client is initialized with `vertexai: true`. This ensures requests are billed to your Google Cloud credits.
- **Project ID** — If the SDK cannot detect your project automatically, add `-e GOOGLE_CLOUD_PROJECT=your-project-id`.
#### Example:
```bash
docker run -it \
-p 8080:8080 \
-v "$HOME/.config/gcloud:/root/.config/gcloud" \
-e GOOGLE_APPLICATION_CREDENTIALS=/root/.config/gcloud/application_default_credentials.json \
-e GOOGLE_CLOUD_PROJECT=your-project-id \
your-image-name