The Kryotta blog
Guides and comparisons on working with every AI model — Claude, Gemini, Llama, Mistral, DeepSeek and more — plus what we're building.
I used Compare to test 4 AI models on 95 Takealot return emails: which one per reason
I tested Claude, Gemini, Llama, and DeepSeek on 95 real Takealot return emails during load-shedding chaos. Here's which model handled each return reason best—and what actually got sent to customers.
I used AI for 40 restaurant order queries during load-shedding: checklist for what works
A Cape Town restaurant owner tested Claude and Gemini on 40 real WhatsApp order queries during load-shedding. Here's what AI handled alone, what it botched, and the cost per response.
I used Auto routing for 140 Takealot return emails during load-shedding: which model it picked per query
I tested AI auto routing on 140 return emails during load-shedding. Here's which model handled each query type and what it cost versus hiring a VA.
I made 45 product Reels for Takealot with Luma and Nova: which style passed review and got clicks
I generated 90 product videos using Luma Ray 2 and Nova Reel to meet Takealot's Black Friday deadline. Here's which AI video style passed review and actually drove clicks.
I made 30 product shots for Takealot with Stability AI during load-shedding: which prompts passed review
I generated 30 product shots with Stability AI during load-shedding to meet a Takealot deadline. Twenty-three passed review. Here's what the prompts needed to work.
I used Claude vs Gemini to process 90 Yoco settlement reports during load-shedding
I tested Claude and Gemini to process 90 days of payment settlement reports during Cape Town's load-shedding crisis. Here's which AI model handled interruptions, deduplication, and reconciliation better.
Claude vs Gemini for Yoco and SnapScan payment reconciliation: I tested 70 reports
I tested Claude and Gemini on 70 real payment reports from my Cape Town coffee roastery. Here's which AI catches duplicate transactions, handles Afrikaans notes, and actually explains discrepancies to non-technical staff.
I used Auto routing for 200 Takealot return queries: which model it picked per task
I tested Auto routing on 200 return queries across my Takealot shop. Here's which model it picked for policy questions, refund instructions, and angry customers—and whether it actually saved time.
Open-weight vs closed AI models: which one for your Takealot store (and when to switch)
Open-weight models like Llama cost less and offer privacy options. Closed models like Claude are more reliable for nuanced tasks. Here's how to pick the right one for your Takealot store—and when to switch.
5 myths about AI product photos and Reels that waste your ad budget
Most businesses believe AI-generated product images look fake and AI video takes forever. We tested both myths with real Shopify and Instagram ads—here's what actually wastes your budget.
5 myths about AI automation that cost SA online sellers time and money
Three myths about AI automation are costing SA online sellers time and money. Here's what actually works: local payment handling, auto-routing, and workflow setup—without the agency markup.