Building an AI Vedic Astrology App in 25 Days — What Actually Worked (and What Didn't)

25 days ago I had an idea. Today KundliAI is live, indexed by Google, and getting organic traffic.

Here's the honest story — no fluff.

The Idea

India has 1.4 billion people. A massive percentage consult astrology regularly — for marriage, career decisions, naming babies, starting businesses.

But the existing platforms are either:

Outdated UI from 2005
English-heavy (most users are Hindi speakers)
Expensive (₹500-2000 per consultation)
Generic (same horoscope for everyone born in the same month)

I thought — what if I built something that gives genuinely personalized Vedic astrology in Hindi, free, in 30 seconds?

That became KundliAI.

The Stack

Frontend: Next.js 14 (App Router)
Styling: Tailwind CSS
AI: Groq API (Meta's Llama — fast and cheap)
Astrology calculations: AstrologyAPI.com (Swiss Ephemeris)
Payments: Razorpay
Hosting: Vercel
Domain: kundliai.in

Why Groq over OpenAI? Speed. Groq is 10x faster for inference. For a tool where users expect results in 30 seconds — latency matters a lot.

Why AstrologyAPI over doing calculations myself? Swiss Ephemeris is complex. AstrologyAPI wraps it cleanly and gives me planetary positions, Kundli Matching (36 Guna Milan), Mahadasha calculations — all via REST API. Cost is minimal for the accuracy you get.

What I Built — Features

Free Kundli Generator
User enters name, date of birth, time, place. We call AstrologyAPI for planetary positions, then Groq generates a complete personalized reading in Hindi. Takes about 25-30 seconds end to end.

Daily Rashifal (Horoscope)
Not generic — based on actual planetary transits for the day combined with user's Moon sign. Different from every other rashifal site that gives the same text to all Aries users.

Kundli Matching
36 Guna Milan — the traditional Indian marriage compatibility system. Enter two birth charts, get compatibility score with explanation in Hindi.

Numerology
Name + date of birth → Life Path Number, Destiny Number, Soul Urge Number. Hindi interpretation.

Love Compatibility by Name
Chaldean numerology — enter two names, get compatibility score. Simple but surprisingly popular.

AI Pandit Chat
24/7 chat with an AI that knows Vedic astrology. Ask anything — career, marriage, health, planetary positions.

Panchang
Daily Panchang for 12 Indian cities — Tithi, Nakshatra, Yoga, Rahu Kaal, Abhijit Muhurat.

Paid Reports (₹99)
Detailed 10-page PDF report. One-time payment via Razorpay. First payment came on Day 20.

The Technical Challenges

Challenge 1 — Prompt Engineering for Hindi

Getting Groq/Llama to generate genuinely good Hindi astrology content was harder than expected.

The model kept mixing Hindi and English awkwardly. It would use Sanskrit terms without explaining them. The tone was sometimes too formal, sometimes too casual.

After about 40 iterations of the system prompt, I landed on something that works. Key insights:

Give the model a persona ("You are an experienced Vedic astrologer who explains things in simple Hindi")
Specify the audience ("The user is a regular Indian, not an astrology expert")
Give explicit formatting instructions
Include examples of good vs bad responses in the prompt

Challenge 2 — Vercel Serverless File Size Limits

My blogs.ts file grew to 500KB+ with 32 blog posts. Vercel's serverless functions have memory limits and the file was causing cold start issues — only 3 blogs were loading.

Solution: Split into blogs.ts (9 blogs) and blogs2.ts (23 blogs), import both in the page component.

Lesson: Watch your serverless function bundle size. It bites you in production, not in development.

Challenge 3 — SSR vs Client Rendering for SEO

Initially blog pages were client-rendered. Google was seeing empty pages — terrible for SEO.

Migrated to Server Components with force-dynamic. Now Google gets fully rendered HTML with all content. Indexed 2 new blogs in under 10 minutes after this fix.

Challenge 4 — AstrologyAPI Rate Limits

During testing I burned through API credits quickly. Built a simple caching layer — same birth data within 24 hours returns cached results instead of hitting the API again.

Cost dropped significantly. API cost is now ₹8-17 per day even with decent traffic.

The SEO Strategy — What's Working

I'm not an SEO expert. But here's what I did:

Blog content in Hindi + English
Most astrology content online is either pure English or poor Hindi. I write detailed bilingual blogs — "Makar Rashi 2026 rashifal" with proper Hindi content. Google seems to like this.

Long-tail keywords
Not "horoscope" (impossible to rank). Instead "makar rashi sade sati 2026 upay" — specific, lower competition, high intent.

FAQ sections on every blog
Google's featured snippets love FAQ format. I include 10-14 FAQ questions per blog with direct answers.

Schema markup
FAQ schema, Breadcrumb schema, Review schema — all implemented server-side. Google Search Console shows all valid. Rich results in search = higher CTR.

Results after 25 days:

0 to 1,390 impressions
67 clicks
4.8% CTR (industry average is 2-3%)
Position 7.6 average
First paid customer on Day 20

The Distribution Strategy

SEO takes time. Meanwhile:

Quora — answering astrology questions with genuine helpful content. 51 answers so far. 166 content views this month. Referral traffic is steady.

Featured.com — expert Q&A platform. Submitted 3 answers. One selected by BoostmyDomain for publication. Two in review with Brodox and VisRank. High-DR backlinks incoming.

IndieHackers — sharing the journey. Good for backlinks and community feedback.

Revenue — Honest Numbers

Day 20: First payment — ₹99
Day 25: Total revenue — ₹99-200

Not impressive yet. But the paid report feature works, Razorpay integration is solid, and the funnel is there. It's a traffic problem right now, not a conversion problem.

What I'd Do Differently

Start SEO content on Day 1
I built features for the first 2 weeks before writing any blogs. Those 2 weeks of content compound over time — I wish I'd started earlier.

Pick a narrower niche first
"AI Vedic astrology" is broad. Starting with "Sade Sati calculator" or "Kundli Matching tool" and dominating that before expanding would have been smarter.

Build in public from Day 1
I started sharing on Day 20. Earlier sharing = earlier feedback = earlier course corrections.