The Problem 🤯
If you've ever tried to open a demat account in India, you know the pain — PAN verification, Aadhaar e-KYC, DigiLocker consent, bank penny drop, KRA registration, CKYC upload, eSign... the list goes on.
Now imagine you're the engineer who has to build that system. You're staring at 25+ vendor APIs, 8 regulatory bodies, 454 data fields, and SEBI circulars that change every quarter.
I needed a single source of truth. Something that connects the dots between what the user sees on screen and what happens behind the scenes — the API calls, the batch pipelines, the maker-checker workflows, the exchange registrations.
So I built one. With Claude Code as my co-pilot. 🤖
What I Built 📚
KYC Onboarding Specification — a 57-page technical docs site covering:
- 🖥️ 9-screen mobile-first user journey — from mobile OTP to eSign in ~6 minutes
- 🔌 25+ vendor integrations — Decentro, Digio, HyperVerge, Leegality, TrackWizz, and more
- 🏭 8 parallel batch pipelines — KRA, CKYC, NSE, BSE, MCX, CDSL, NSDL, Income
- ✅ Maker-checker admin workflow — because compliance teams don't trust automation (rightly so)
- 📋 454-field master dataset — every field mapped to its source, validation rule, and destination
- 📜 Full SEBI/KRA/CKYC regulatory mapping — current as of Feb 2026
How Claude Code Helped 🛠️
I didn't write this in Google Docs. I used Claude Code (Anthropic's CLI agent) to research, draft, and iterate on the entire spec.
Here's what that looked like in practice:
Research phase — I'd ask Claude to deep-dive into vendor APIs. "Read the Decentro docs and map their PAN verification response to our field schema." It would search, read, cross-reference, and produce structured tables.
Drafting — Each of the 57 pages went through Claude. I'd describe the screen flow or pipeline step, and Claude would draft the technical spec — field mappings, API sequences, error handling, edge cases. I'd review, correct, iterate.
Regulatory cross-referencing — SEBI circulars are dense. I'd feed Claude the circular text and ask "What does this mean for our KRA upload format?" Getting structured answers saved hours of lawyer-brain parsing.
The docs site itself — Built with Astro Starlight. Claude scaffolded the config, created custom components (screen cards, vendor badges, async bars, status flows), wrote the CSS, and even generated the OG image programmatically.
SEO & distribution — Claude set up JSON-LD structured data, robots.txt, llms.txt, IndexNow submissions to 4 search engines, Wayback Machine archival, and Cloudflare analytics. All from the terminal.
The Tech Stack 🏗️
| Layer | Tool |
|---|---|
| Docs framework | Astro Starlight |
| Hosting | GitHub Pages |
| AI co-pilot | Claude Code (Opus) |
| Analytics | Cloudflare Web Analytics |
| Search indexing | IndexNow + GSC + Bing + Yandex |
| Diagrams | Hand-crafted SVG |
Key Takeaway 💡
Building compliance systems is 20% code and 80% understanding the rules. The hardest part isn't the API integration — it's knowing that CKYC search now returns masked numbers (since Jan 2025), or that DDPI replaced POA (since Nov 2022), or that nominee opt-out requires video verification (since Jan 2025).
An AI co-pilot doesn't replace domain expertise. But it compresses the research-to-documentation cycle from weeks to days. I went from "we need a KYC spec" to a fully published 57-page docs site in a fraction of the time it would have taken solo.
Check It Out 👀
🔗 Live site: javajack.github.io/broking-kyc
🐙 GitHub: github.com/javajack/broking-kyc
If you're building fintech in India — especially onboarding, KYC, or compliance systems — I hope this saves you some pain. PRs and feedback welcome.
I'm Rakesh Waghela, a Tech & KYC Solutions Architect. If you're building KYC or compliance systems and want to chat, find me on LinkedIn, X (@webiyo), or book a call on Topmate.
Top comments (0)