Twelve months ago, we started building VIZIQO with a simple thesis: Indian businesses need AI-powered tools built for Indian workflows, at Indian price points. Today, we have two live products — VIZIQO Assist (AI call center) and VIZIQO Campaign (marketing automation) — serving paying customers. Here's what we actually learned along the way.
Choosing the Stack: Optimize for Speed, Not Perfection
We spent exactly one week on stack selection — and we're glad we didn't spend more. For Assist: FastAPI + Python (because the AI/ML ecosystem is Python-first), Qdrant for vector search, LangChain for orchestration, and Next.js for the dashboard. For Campaign: NestJS + TypeScript for the API, PostgreSQL + Prisma, and Next.js again for the frontend.
The lesson: pick technologies your team knows well, that have strong ecosystems, and that let you ship fast. We considered Rust for performance-critical parts of Assist, but Python's AI library ecosystem saved us months of development time. Premature optimization would have killed our timeline.
Payments in India: Razorpay Wins, But Plan for Edge Cases
Razorpay was the obvious choice for payment processing, and it's been solid. But the edge cases will surprise you: GST invoice generation requirements, handling subscription pauses (not just cancellations), managing failed recurring payments gracefully, and dealing with RBI's auto-debit regulations for recurring payments above ₹5,000.
Budget at least 2 weeks for payment integration, not the 2 days the documentation suggests. The core integration is fast — it's the compliance and edge case handling that takes time.
Compliance Is Not Optional
DLT registration for SMS, WhatsApp Business API approval, data localization requirements, GST compliance, terms of service, privacy policy — the compliance overhead for B2B SaaS in India is real. We spent roughly 15% of our first six months on compliance-related work.
Our advice: start compliance processes on day one, in parallel with development. DLT registration alone can take 2-4 weeks. WhatsApp Business API approval requires a verified Facebook Business Manager account, which has its own timeline.
Infrastructure: Start Simple, Scale When You Must
We launched on a single EC2 instance. Not a Kubernetes cluster, not a multi-region setup — one t3.xlarge instance running Docker Compose. Total infrastructure cost for the first three months: under ₹15,000/month. When we hit scaling issues (around 200 concurrent users on Assist), we moved to ECS with auto-scaling. Still not Kubernetes — ECS is simpler and sufficient for our scale.
The lesson: your infrastructure should match your current scale, not your aspirational scale. Every hour spent on infrastructure that isn't needed yet is an hour not spent on features your customers are asking for.
What We'd Do Differently
Start charging from day one — we offered a free beta for too long and attracted users who were never going to pay. Build admin tooling earlier — we spent too many hours manually handling things that a simple admin dashboard could automate. And invest in monitoring from the start — we were flying blind for the first two months, and it cost us in downtime and debugging time.
Building SaaS in India is challenging but entirely viable. The market is growing, customers are willing to pay for good products, and the infrastructure ecosystem is mature enough to support serious applications. The key is staying focused, shipping fast, and not over-engineering.