AI developer story chapters are usually told backwards from success, so let me tell mine forwards from confusion: a Junagadh boy, a first ugly website in 2018, and a slow climb to shipping AI systems at 62 tok/s on fixed ₹55K builds.
I am Deepak Bagada, founder of SaaS Next in Junagadh, Gujarat. No funding announcements, no overnight wins — just six chapters of compounding skills where each one paid for the next. This is the honest version, with the invoices and the mistakes attached.
Chapter One (2018): The Ugly First Website
I bought my first domain in 2018 and built a website so slow it barely loaded on hostel Wi-Fi. It had flashing text, three fonts, and a hit counter. I loved it completely. That site taught me the only lesson that matters at the start: shipped and ugly beats perfect and hypothetical.
I wrote HTML by day and broke it by night. View-source was my university. Within a year I could build anything a small shop asked for — badly, but completely. My web development services still carry that chapter-one DNA: a real page on a real URL beats a mockup every time.
<!-- File: index-2018.html — my actual skill level, preserved without shame -->
<!DOCTYPE html>
<html>
<head><title>My First Site</title></head>
<body>
<h1>Welcome to my website</h1>
<p>I build websites in Junagadh. Call me.</p>
</body>
</html>
(Angle brackets escaped in print so publishing tooling reads the file cleanly; the code is exactly what it looks like.)
Chapter Two (2020): Marketing Hurt My Pride
By 2020 I could build solid sites that nobody visited. A client asked why his beautiful store got eleven visitors a month, and I had no answer. That silence stung more than any bug. So I learned SEO, content, and analytics — slowly, skeptically, one ranking at a time.
The unfair advantage appeared fast: developers who understand search are rare in small markets. My sites started ranking, clients started calling, and I stopped competing on hourly rates. The full playbook from that era now lives in my traffic tripling case study, where the same habits took SaaS Next from 18K to 55K sessions.
War Story 1: The ₹8,000 Site That Taught Pricing (2021)
Chapter three opened with my first paid Laravel website in 2021: ₹8,000 for a complete business site, two weeks of evenings, three revision rounds that tripled the scope. Someone I had never met paid for something I made, and the feeling genuinely never wore off. But the math did bother me — roughly ₹110 an hour after revisions.
That project forced fixed-scope contracts with one revision round and a price list I stopped apologizing for. Next identical site: ₹25,000, one week, happier client. The lesson touring founders learn late: pricing is a product decision, not a reward for suffering. Today my fixed builds run ₹55K–₹85K, and the scope document from that ₹8,000 job is still the ancestor of every quote I send.
Chapter Four (2023): The Unscripted Answer
In 2023 I asked a large language model a question I had not scripted, and it answered correctly from documents I had given it. I remember the chair I sat in. Every previous tool I had mastered did exactly what it was told; this one read and reasoned. I knew the next decade of my work would be about this technology.
I spent that year rebuilding: Python beside PHP, embeddings beside indexes, eval sets beside test suites. Evenings after client work, my Junagadh desk became a lab with two laptops and one very patient ceiling fan.
War Story 2: The Demo That Failed in Front of Everyone (2024)
Chapter five nearly ended early. In 2024 I demoed my first LLM product — an early Curro prototype — live to a Surat client over screen share. Mid-demo the model invented a confident, completely wrong price list, latency spiked to 9 seconds per answer, and the client laughed in the worst possible way. I wanted the floor to open.
That humiliation built my current stack: RAG with citations over pgvector, eval sets that must pass before any demo, and a latency budget of P95 under 2 seconds that kills features on sight. Curro 1.0 shipped months later on those rules, documented in Curro 1.0, my AI content studio, and it now drafts at 62 tok/s without inventing prices. Failure in public is expensive tuition with excellent retention. My AI development services inherit every scar from that afternoon.
# File: pre-demo.sh — the gate that demo-day disaster bought me
php artisan test --filter=EvalSuite
node evals/run-heldout.mjs --min-accuracy 90 --max-p95-ms 2000
# Demo ships only when both gates pass. No exceptions, no "it worked yesterday".
Chapter Six (Today): The Compounding Desk
Today the chapters stack instead of competing. Code ships the product, marketing compounds its reach, and AI multiplies both — one desk in Junagadh, clients across Gujarat, systems that answer while I sleep. The backend habits keeping it all upright are written up in my Laravel production lessons: queues offline, migrations boring, deploys health-gated.
| Chapter | Year | Skill added | What it pays today |
|---|---|---|---|
| Ugly first site | 2018 | Shipping without permission | Speed over polish instinct |
| Marketing pride-hit | 2020 | SEO + analytics | Rankings that compound |
| First paid Laravel job | 2021 | Fixed-scope pricing | ₹55K–₹85K builds |
| Unscripted answer | 2023 | LLMs + evals | 62 tok/s assistants |
| Curro demo disaster | 2024 | RAG + latency budgets | Products that do not lie |
| Compounding desk | 2026 | All of it at once | One-person full stack |
When NOT to Copy My Path
Do not copy my timeline if you need income in ninety days — learn one billable skill deeply instead of my scenic route through three fields. Do not copy my self-taught gaps either: I skipped computer-science fundamentals and paid for it in wasted months on algorithms a textbook covers in a weekend. And do not confuse my constraints with virtues. Building from Junagadh taught me frugality, but proximity to mentors and markets is genuinely useful, and I travel for it now. Take the compounding mindset, skip the detours I have marked for you.
Frequently Asked Questions
What is your AI developer story in brief?
My AI developer story runs six chapters from a first ugly website in 2018 to shipping AI systems from Junagadh in 2026. Code paid for marketing skills, marketing paid for clients, and both now compound inside AI products built at fixed ₹55K–₹85K prices.
How long does it take to go from code to AI work?
My path took five years of evenings, but a focused developer today can compress it: one year of shipping web apps, six months of SEO and analytics, then a year of LLMs with evals. Building public projects each month matters more than certificates.
Can you build an AI career from a small city in India?
Yes. I run everything from Junagadh, Gujarat with clients across the state, holding P95 210ms on a ₹1,850/month stack. Small-city costs are the advantage: low overhead lets you price fixed builds that metros cannot touch.
What should I learn first — code, marketing, or AI?
Code first, because shipping is the foundation everything else multiplies. Then enough marketing to get your work seen, then AI on top. Skipping straight to AI without shipping habits produces demos that fail in public, as my 2024 screen-share proved.
Bottom Line
AI developer story, six chapters, one lesson: ship ugly in 2018, learn search in 2020, price honestly in 2021 — then let AI multiply all three.