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 made 50 product Reels for Instagram Shop with Luma and Nova: which style got saves
I tested Luma Ray 2, Nova Reel, and Stability's image-to-video on 50 product Reels for my Shopify store. Slow camera moves won on saves—here's what the data showed.
5 myths about AI in restaurants and healthcare that waste your automation budget
Most businesses waste their AI budget chasing flashy use cases. The real ROI comes from automating unglamorous admin work—reservation confirmations, vendor emails, intake forms—that costs money when done manually.
I used Auto routing for 220 eBay and Shopify returns: which model it picked per query
I processed 220 return requests over six weeks using Auto routing to pick the best AI model for each query. Here's what it chose and why it mattered for my small business.
I used Auto routing for 180 Allegro and Bol return emails: which model it picked per query
I processed 180 return emails from Allegro and Bol.com using AI auto routing. Here's which model it picked for each type of request and whether those choices actually saved time.
I made 40 product shots for Shopify with Stability AI — which prompts passed and which got refunds
I generated 40 product images with Stability AI for my ceramic shop. Some sold immediately. Others triggered refunds. Here's which prompts customers actually trust.
I used Claude vs Gemini to write 150 GST invoice reminders: which one customers paid faster
I sent 150 overdue GST invoice reminders split between Claude and Gemini to test which AI writes reminders that actually get paid faster. Claude's reminders got paid 40% faster—6.8 days vs 11.4 days.
I explained tokens, context windows and grounding to my Nairobi team — what finally made sense
A Nairobi team struggled to understand AI costs and capabilities until we compared tokens to M-Pesa records and context windows to WhatsApp chat history. Here's what finally clicked.
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.
I used Claude vs Gemini to reply to 120 Flutterwave dispute emails: which model won
I tested Claude and Gemini on 120 Flutterwave dispute responses. Claude won 9 out of 11 cases. Here's what separates a winning response from one that gets rejected.
I used 5 prompt templates to reply to 300 Flipkart and Meesho vendor messages: which formula per task
I tested five AI prompt templates on 300 vendor messages across Flipkart, Meesho, and WhatsApp. Here's which formula worked best for order confirmations, delays, pricing, invoices, and follow-ups.
OpenAI o1 vs Claude 3.7 Sonnet for Xero invoice reconciliation: I tested 60 reports
I tested OpenAI o1 and Claude 3.7 Sonnet on real invoice reconciliation work—60 transactions, VAT checks, duplicate detection, and missing receipts. Here's what actually worked.
I used AI to process 80 Companies House sole trader invoices in one afternoon: step by step
I tested Claude and Gemini on 80 sole trader invoices: structuring data, verifying Companies House records, and calculating UK VAT. Here's what worked, what failed, and what I'd do differently.