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 4 prompt templates to reply to 250 WhatsApp customer orders: which formula per task
I tested 4 prompt templates across 250 WhatsApp customer orders over three weeks. Here's which formula reduced follow-ups, sped up replies, and made my shop look organised.
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 used Auto routing for 190 Zalando and Bol return emails: which model it picked per query
I processed 190 return emails across Zalando and Bol using AI auto routing. Here's which model the system picked for each query type, and whether it actually saved time.
I used 4 prompt templates to write 180 eBay listing descriptions: which formula per category
I tested four prompt templates across Claude, Gemini and Llama to write 180 eBay listing descriptions in two weeks. Here's which formula and model worked best for electronics, vintage fashion, collectibles and home goods.
I used AI in 6 industries for 90 days: which tasks it finished vs which needed me
I tested AI across six UK businesses for 90 days. Certain tasks—customer replies, invoices, first drafts—came back ready to send. Others, like VAT reconciliation and judgment calls, needed me every time.
I used 3 AI models to write 80 Cyber Monday email sequences: which one per open rate goal
I tested Claude, Gemini, and GPT on 80 Cyber Monday emails across three segments. One model dominated abandoned-cart recovery. Another crushed VIP messaging. Here's which AI wins for each goal—and why the prompt matters more than the model.
I used 5 explainer prompts to teach my team tokens and context windows: which one clicked
I built five AI prompts to teach my team about tokens and context windows. Three worked perfectly. Here's which explanation finally made the concepts click—and what I'd do differently.
I explained tokens, context windows and hallucinations to my Meesho vendor team — what finally made sense
I tried teaching my vendor team how AI actually works—tokens, context windows, hallucinations—using their own work. Some analogies flopped. Here's what finally clicked.
I used 4 AI models to write 100 GDPR data requests in one week: which one per case
I cleared a 103-request GDPR backlog in one week using Claude, Gemini, DeepSeek, and GPT-OSS. Here's which model handled each request type fastest and why it cost only €8.30.
I used 5 prompt templates to write 200 Shopify refund emails: which formula per situation
I handled 200 refund requests in one week using five AI prompt templates. Here's which formula I used for each situation, the exact prompts, and why switching between Claude and Gemini kept my approval rate at 30%.
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.
5 myths about new reasoning models that waste your Razorpay automation budget
Reasoning models aren't always the answer for Razorpay automation. Learn which tasks actually need them—and which ones waste your budget.