I Built an AI Tool That Rips Apart Business Ideas in 60 Seconds. Here’s How (and Why)
- Mar 14
- 6 min read
A brutally honest, 7-phase validation engine for Indian founders — powered by Claude, built in a weekend.
Every founder thinks their idea is the one.
I’ve sat through pitch decks that open with “India has 1.4 billion people” as if that alone justifies building yet another hyperlocal delivery app. I’ve watched smart people burn through 50 lakhs validating assumptions they could have killed with a Google search and a spreadsheet.
The problem isn’t ambition. It’s that nobody stress-tests the idea before writing the first cheque.
So I built a tool that does exactly that. No sugarcoating. No “interesting concept, but…” hedging. Just a cold, structured breakdown of whether your business idea has legs — or is walking into a graveyard.
Here’s how it works, why I designed it this way, and how you can build your own.
The Problem: Founders Validate Too Late
Most Indian startup founders I’ve spoken with follow the same pattern:
Step 1: Get excited about an idea.
Step 2: Build an MVP for 3–6 months.
Step 3: Launch. Get 200 users. Realise the unit economics don’t work.
Step 4: Pivot. Repeat.
The real validation — TAM sizing, competitor mapping, unit economics, AI obsolescence risk — happens after money is spent. Sometimes it never happens at all.
I wanted a tool that forces you to confront these questions on Day 0. Before you quit your job. Before you hire a co-founder. Before you touch a single line of code.
What the Tool Actually Does
You paste in your raw business idea. The tool runs it through 7 phases of analysis and returns a structured report in about 60 seconds.
Here are the 7 phases:
Phase 1: The Brutal Deconstruction Strips your idea to one sentence. Identifies the real friction you’re ignoring. Lists the 3 ways people already solve this problem without you.
Phase 2: India Market & Bloodbath Analysis Calculates TAM in Crores with a logical breakdown. Names 3 incumbents. Finds startups that tried this and died — and explains exactly why they failed.
Phase 3: Global Context & Winners Finds international companies running this model successfully. Then explains why the same model might fail in India (willingness to pay, trust gaps, infrastructure).
Phase 4: Unit Economics & Capital Math Breaks down the money engine: how you make money, what CAC and LTV look like, and what it costs to scale to ₹100 Crores.
Phase 5: AI Threat Matrix & Social Sentiment The phase most founders skip entirely. Could an AI model make your entire business irrelevant in 18 months? What are real people saying about this problem on Twitter, Reddit, and LinkedIn?
Phase 6: Execution Reality & Timelines Top 3 assumptions you need to prove in 90 days. The exact 3 hires you need on Day 1 (not “Sales Head” — more like “B2B closer with a pharma network”). Best case and worst case scenarios with actual numbers.
Phase 7: The Final Verdict A definitive GO or NO-GO. Two sentences. No hedging.
The Design Philosophy: Scepticism as a Feature
The system prompt behind this tool is opinionated by design. A few key choices:
Scepticism-first thinking. The AI assumes your idea is flawed and works backwards. This is the opposite of how most people use ChatGPT (“Tell me why my idea is great”). Founders need someone who will find the holes, not confirm the bias.
India-first context. All financials use Rupees, Lakhs, and Crores. Competitors are Indian companies. Market sizing is India-specific. Global parallels are included, but always with an “adaptation risk” section explaining why copy-paste from the US market rarely works here.
Web search is mandatory. The tool uses Claude’s web search capability to pull live market data, find real competitors, and scan social media sentiment. No stale training data. Every report is grounded in what’s actually happening right now.
Forbidden words. The prompt explicitly bans “synergy,” “disruptive,” “game-changer,” and “innovative.” If your analysis needs those words to sound convincing, the analysis is weak.
The Technical Build: Simpler Than You Think
The entire app is three files that matter:
1. The Backend (server.js)
A minimal Express server that does one thing: proxy API calls to Anthropic so your API key stays on the server.
app.post("/api/validate", async (req, res) => {
const response = await fetch("https://api.anthropic.com/v1/messages", {
method: "POST",
headers: {
"Content-Type": "application/json",
"x-api-key": process.env.ANTHROPIC_API_KEY,
"anthropic-version": "2023-06-01",
},
body: JSON.stringify(req.body),
});
const data = await response.json();
res.json(data);
});
That’s the entire API layer. No database. No auth. No middleware stack. Just a proxy.
2. The System Prompt
This is where all the intelligence lives. The prompt is ~800 words of very specific instructions that force the AI into a structured output format. The key technique is output architecture — telling the model exactly which sections to produce, in what order, with what content.
The critical line in the API call:
tools: [{ type: "web_search_20250305", name: "web_search" }]This enables Claude’s built-in web search, which means the analysis includes live data — real competitor names, real market numbers, real social media sentiment.
3. The Frontend (BusinessValidator.jsx)
A single React component. No routing. No state management library. Just useState and useEffect.
The UI has three stages: input (textarea + example ideas), loading (spinner + phase tracker), and result (rendered markdown). A simple markdown parser converts the API response into styled HTML.
One nice touch: the phase tracker at the top lights up as each phase appears in the response, so you can watch the analysis build in real-time.
Total lines of code: Under 500. Time to build: One weekend. Dependencies: React, Express, cors. That’s it.
What I Learned Building This
Prompt engineering is product design
The system prompt is the product. The React app is just a delivery mechanism. I spent more time refining the 7-phase structure than writing any code. Which phases to include, what order they should appear in, how to force the AI to be specific rather than generic — these are product decisions, not engineering ones.
Web search changes everything
Without web search, the tool would produce generic analysis based on training data. With it, you get actual competitor names, actual market numbers, and actual social media quotes. The difference between “there are several competitors in this space” and “Shiprocket, GoKwik, and Logistiq are already doing this” is the difference between useless and useful.
Constraints make better outputs
The “forbidden words” list sounds trivial, but it dramatically improves output quality. When you ban buzzwords, the AI is forced to be specific. Instead of “this is a disruptive approach to the market,” you get “this undercuts incumbent pricing by 40% but requires 3x the operational complexity.”
The graveyard section is the most valuable
Phase 2 includes a “Graveyard” section — startups that tried this and died. This is consistently the most eye-opening part of any report. Founders rarely research who failed before them and why. The AI does it automatically with web search.
Try It Yourself
The full source code is available as a ready-to-run project. Here’s the quickstart:
# Clone or unzip the project
cd business-validator
npm install
npm install express cors
# Set your Anthropic API key
export ANTHROPIC_API_KEY=sk-ant-your-key-here
# Run both frontend and backend
npm run dev
Browser opens at localhost:3000. Paste in any business idea. Wait 60 seconds. Read the truth.
You’ll need an Anthropic API key from console.anthropic.com. Each validation costs roughly $0.05–0.10 in API credits depending on the length of the response and number of web searches.
What’s Next
A few things I’m considering:
Comparison mode. Paste in two competing ideas and get a head-to-head analysis — which one has better unit economics, a larger TAM, and fewer incumbents.
PDF export. One-click download of the full report as a PDF you can share with co-founders or investors.
Industry presets. Pre-loaded prompts for specific verticals (fintech, healthtech, edtech, D2C) that ask industry-specific questions.
Follow-up questions. After the initial report, let the user drill deeper: “Tell me more about the graveyard companies” or “How would the unit economics change at 10x scale?”
The Uncomfortable Truth
Most business ideas don’t deserve 5 years of your life.
That’s not pessimism — it’s math. The TAM is too small, the incumbents are too strong, the unit economics don’t work, or an AI model will do it for free in 18 months. Better to know on Day 0 than Day 900.
This tool won’t tell you what to build. But it will tell you — with cold, data-backed specificity — whether what you’re about to build is worth building.
Sometimes the most valuable thing a founder can hear is “No.”
If you’re building in the Indian startup ecosystem and want to stress-test your idea, the tool is open source. Fork it. Break it. Make it better. The system prompt alone is worth stealing.
Have questions about the build or want to discuss the architecture? Drop a comment or find me on Twitter.
Comments