Blog

Engineering Notes, Written From Production

Voice AI latency, SaaS architecture, and how software actually gets shipped — from a team that operates its own product, not just other people's.

Voice AI6 min read

Engineering ~0.7s Voice Response in a Production Voice Agent

A stage-by-stage teardown of how TeleVox AI hits ~0.7s voice-to-voice latency: VAD, streaming STT, LLM token streaming, early TTS, and barge-in.

Read
AI7 min read

RAG That Answers Correctly: Building Business Knowledge Bases

How we build RAG that answers correctly: structure-aware chunking, pgvector in Postgres, the retrieval failures that bite, and evals before shipping.

Read
Mobile6 min read

React Native vs Flutter in 2026: How We Actually Choose

We have shipped React Native to trading platforms and Flutter into healthcare IoT. The four questions that decide the choice, plus a decision table.

Read
Engineering6 min read

MVP Architecture That Survives to Series A

The week-one architecture decisions that are expensive to reverse — data model, tenancy, auth, billing hooks, deploys — and what to deliberately keep scrappy.

Read
Process5 min read

Why We Only Do Fixed-Price Contracts (and How We Survive It)

Hourly billing rewards slow delivery. How we make fixed-price contracts work — paid discovery, one-week sprints, Friday demos — and when the model breaks.

Read
Voice AI6 min read

What Voice AI Can Actually Do for a Local Business in 2026

What AI receptionists reliably handle for local businesses in 2026, where they still fail, and the questions to ask any voice AI vendor before you buy.

Read

Ready to Build Something That Lasts?

Wherever you're based — book a 30-minute call to talk through your product, timeline, and how KiswahTech can help.

30 minutes · Response within 2 hours · Founder-to-founder