Thinking out loud
What we're building, learning, and shipping at OptiAI Tech — written for founders and SMB owners.
Why we chose Claude API over GPT-4 for our WhatsApp agent (and what we learned)
After running both models in production for 3 months, here's the honest comparison — accuracy, cost, hallucination rate, and which one Indian SMB owners actually get better responses from.
// production stats · last 30 days
model = "claude-sonnet"
hallucination_rate = 0.018
avg_response_ms = 1840
messages_processed = 1_247_903
businesses_active = 500
Apr 2025
Building Opswake: from a frustrating afternoon to paying SMB customers in 6 months
The honest story of how Opswake was built — the wrong assumptions, the pivots, the moment it clicked, and what Indian SMB owners actually needed versus what we thought they needed.
Apr 2025
How we handle 50,000 WhatsApp messages a day without dropping one
The webhook architecture, queue system, and failure handling that keeps Opswake's AI responding in under 2 seconds even during peak load. Practical lessons for anyone building on the Meta Cloud API.
Mar 2025
Why every Indian SMB tool needs to be WhatsApp-first (not just mobile-first)
Mobile-first has been the mantra for a decade. For Indian SMBs in 2025, that's not enough. The real insight is WhatsApp-first — and it changes how you design everything from onboarding to notifications to data entry.
Mar 2025
Why most custom software projects fail (and the 3 questions we ask before taking any project)
70% of custom software projects either fail, run massively over budget, or deliver something nobody uses. After building products and custom solutions for dozens of businesses, here's what we've learned.
Feb 2025
The Indian SMB AI stack in 2025: what's actually worth using
There's no shortage of AI tools. There is a shortage of AI tools that work for a coaching institute in Pune or a travel agency in Ahmedabad. Here's what's actually useful, what's overhyped, and what we'd recommend.