VidSaar

Hey IH 👋

I'm Sunny, a non-technical solo founder from Maharashtra, India. I just launched VidSaar (vidsaar.com) — an AI tool that turns YouTube lectures into summaries, flashcards, and quizzes in Hindi, Hinglish, and 18 other languages.

I built the entire thing using Claude Code (Anthropic's AI coding tool). No co-founder. No freelancers. No coding bootcamp. Just me describing features in plain English and AI writing the code.

Here's the full breakdown of what I built, how I built it, and what I learned.

## The Problem

India has 50M+ students preparing for competitive exams (JEE, NEET, UPSC). Most of them study from YouTube channels like Physics Wallah (13.5M subscribers), which post 2-3 hour lectures in Hindi.

The problem: students watch for hours and remember nothing. Every AI summarizer on the market (NotebookLM, NoteGPT, Eightify) is English-only, priced in USD, and has zero awareness of Indian exam patterns.

I kept thinking, someone should build this for Indian students. Then I realized I could be that someone. Even without knowing how to code.

## What I Built

VidSaar converts any YouTube video into:

- Structured summaries (in 20 languages, including Hindi and Hinglish)

- AI-generated flashcards with spaced repetition (SM-2 algorithm)

- Multiple-choice quizzes matching actual exam patterns (JEE, NEET, UPSC)

- An AI doubt solver that answers follow-up questions about any video

- Video upload support (up to 2GB, any format)

The product page count got a bit out of hand:

- 38 pages (public + authenticated)

- 46 API routes

- 12+ database tables with Row Level Security

- 134 git commits

## The Tech Stack

- Next.js 14 (App Router), TypeScript, Tailwind CSS

- Supabase (auth, PostgreSQL, storage)

- Anthropic Claude Sonnet 4 (summarization)

- Claude Haiku 4.5 (doubt solver, support chatbot)

- OpenAI Whisper (video upload transcription)

- Razorpay (Indian payments — UPI, cards)

- Paddle (international payments)

- Railway (hosting)

- Cloudflare (CDN, email routing)

## How I Actually Built It (The AI-Assisted Process)

My workflow for every feature:

1. Describe what I want in plain English (a detailed prompt)

2. Paste it into Claude Code

3. Claude writes the code

4. I test it locally

5. Fix issues by describing the bug to Claude

6. Push to Railway (auto-deploys from GitHub)

That's it. Every single page, every API route, every database migration — all generated through prompts.

Some examples of prompts I used:

- "Build a YouTube summarization pipeline that fetches transcripts, chunks long videos, and sends each chunk to Claude for summarization."

- "Add Razorpay checkout with webhook verification, payment deduplication, and atomic token operations."

- "Create a spaced repetition study mode using the SM-2 algorithm."

The key was writing VERY detailed prompts. Vague prompts = vague code. Specific prompts = working features.

## The Security Story

Before launching, I ran a Trail of Bits-style security audit on the entire codebase. Found 52 vulnerabilities. Fixed all of them.

Critical fixes included:

- Atomic SQL functions for token operations (prevents race conditions)

- Webhook payment deduplication (prevents double-crediting)

- Timing-safe HMAC comparison for webhook signatures

- Device fingerprinting for free tier abuse prevention

- Disposable email blocking

- Persistent rate limiting via Supabase

I also set up 31 deny rules in Claude Code to prevent it from ever touching .env files, credentials, or running dangerous commands.

Did security before marketing. Best decision I made.

## The Marketing System (Also Built with AI)

I built a separate project (VidSaar-Marketing) with 9 AI content agents that generate 21 pieces of content daily across 7 platforms:

- 5 tweets (Hinglish + English)

- 5 Instagram posts (carousels + reel scripts)

- 3 Reddit posts (subreddit-specific tone and rules)

- 3 LinkedIn posts (founder journey)

- 5 YouTube Shorts scripts

Plus: blog posts, YouTube full video scripts, paid ad campaigns, and guest post articles for 12 high-DA websites.

All content runs through a humanizer that removes AI-sounding patterns. The Reddit posts match each subreddit's specific rules and tone.

Total cost for content generation: $0/month (runs on Claude Code Max subscription).

## Pricing (Designed for Indian Students)

This was tricky. Indian students are broke. Their monthly hostel mess bill is ₹3,000. Asking them to pay $10/month (₹830) like NoteGPT is a non-starter.

My pricing:

- Free: 3 lifetime summaries (not monthly, prevents cycling abuse)

- Starter pack: ₹149 ($3.99) for 20 summaries

- Pro: ₹499/month ($9.99)

The first summary shows everything unlocked (flashcards, quiz, full features). The second summary blurs the premium features. This creates loss aversion — they KNOW what they're missing.

## SEO Strategy

Built specific pages for LLM and search visibility:

- 3 tool pages (/tools/youtube-summarizer, etc.)

- 4 comparison pages (/compare/vidsaar-vs-notebooklm, etc.)

- 2 resource pages (/resources/what-is-video-summarization, etc.)

- 5 segment pages (/for/college-students, /for/upsc, /for/developers, etc.)

SEO score went from 58 to 85 after:

- Converting landing page from client-rendered to SSR (was invisible to Google)

- Adding Website, SoftwareApplication, FAQPage, BreadcrumbList schemas

- Fixing LCP (removed 950ms blocking animation)

- Adding Google Analytics with Consent Mode v2

## What I'd Do Differently

1. Should have validated with a landing page + waitlist before building the full product. I built too many features before getting a single user.

2. The marketing system was overkill for launch. I should have started posting manually on day 1 instead of spending weeks building an automated content engine.

3. Pricing should have been tested earlier. I went through 3 pricing iterations before landing on the current structure.

## Current Status (Day 1)

- Product: live at vidsaar.com

- Users: 0 (launching today)

- Revenue: ₹0

- Monthly cost: ~₹5,000-10,000 (Railway + Supabase + API costs)

- Funding: ₹0 (bootstrapped)

NEET exam is on May 3 — 31 days away. 24 lakh students are panicking. That's my launch window.

## The Honest Truth

I don't know if this will work. I have zero users, zero revenue, and zero social media followers. But I have a product that genuinely fills a gap; no other tool offers Hindi video summaries with Indian exam modes at ₹149.

The next 90 days will tell me everything. I'm committing to posting content daily, running small Instagram ads (50K INR budget), and tracking every metric.

I'll update this post with real numbers as they come in. Follow along if you want to see whether a non-technical solo founder can build a real SaaS with just AI tools and stubbornness.

## Try It

vidsaar.com — 3 free summaries, no card needed. Paste any YouTube lecture and see what comes out.

Happy to answer any questions about the build process, the AI-assisted development workflow, or the India EdTech market.

— Sunny