Posts

Setting up cost alerts for AI APIs before the bill surprises you

Image
Taking questions during a session Originally published at https://pranjulrathour.scult.in/blog/cost-alerts-for-ai-apis-before-the-bill-surprises-you . That copy is the canonical version and gets updates first. Rate limiting protects your app's stability from a runaway loop. It doesn't protect your wallet — a steady, rate-limited stream of requests can still accumulate an unexpectedly large bill over days if nobody's watching the spend. A minimal cost-alert setup Most LLM providers let you set a spend limit or alert threshold directly in their billing dashboard — set this before you start building, not after the first bill. For a multi-provider setup, track cumulative daily spend in your own logs and alert (even just an email or a log line you check) past a threshold you set. Set the alert threshold meaningfully below your actual budget ceiling, so there's time to react before it's a problem. The scenario this actually prevents A demo that goes unexpectedly...
Post a Comment
Read more

Vision Transformers, explained for practitioners who just want to use one well

Image
Requirements gathering and user flows, on stage Originally published at https://pranjulrathour.scult.in/blog/vision-transformers-explained-for-practitioners . That copy is the canonical version and gets updates first. A Vision Transformer treats an image the way a language model treats a sentence: it splits the image into fixed-size patches, treats each patch like a token, and runs the same self-attention mechanism used in text transformers over those patch tokens. What this practically implies Input resolution and patch size together determine how many tokens the model processes — larger images or smaller patches mean more compute, directly. ViTs generally need more training data than convolutional models to reach the same accuracy from scratch, which is why most practical use starts from a pretrained checkpoint rather than training from zero. Fine-tuning a pretrained ViT on a specific task is usually far more practical for a student project than training one from scratch. Th...
Post a Comment
Read more

Gemini 4 Argon: what Google just launched

Image
Carousel · 9 slides Google announced Gemini 4 Argon on Sep 30, its new frontier model. → Output limit: 1M tokens per answer, up from 64K → Self-reported: 77.9% DeepSWE v1.1, #1 on AutomationBench (51.3%) → Intro price $2 / $10 per 1M tokens, regular $4 / $20 → Rollout: trusted cyber defenders first, then paid API + AI Ultra Slide 7: $2 per million input is now the frontier price at Google, OpenAI, Anthropic and xAI. Sources: Google's announcement and each lab's launch post. I'm Pranjul Rathour, GenAI engineer from Kanpur. Want me to judge your hackathon or speak at your college? pranjulrathour.scult.in/invite Slide 1 of 9 Slide 2 of 9 Slide 3 of 9 Slide 4 of 9 Slide 5 of 9 Slide 6 of 9 Slide 7 of 9 Slide 8 of 9 Slide 9 of 9 Pranjul Rathour GenAI engineer, Kanpur · 3x first-prize hackathon winner · campus mentor I ship production RAG pipelines, fine-tune LLMs and build agentic AI products end to end. I lead engineering at SCULT INDIA for a 14-member team and...
Post a Comment
Read more

Pranjul Rathour: GenAI engineer, campus mentor and speaker. Here is what I bring to your students.

Image
On the mic Originally published at https://pranjulrathour.hashnode.dev/pranjul-rathour-genai-engineer-campus-mentor-speaker . That copy is the canonical version and gets updates first. Most guest speakers show students slides about AI. I show them the systems I shipped this year, the tests that pass, the bills they run up, and the parts that broke. Then we build one together. I'm Pranjul Rathour, a GenAI engineer from Kanpur, India. This page is for the people who decide what happens on a campus: placement officers, heads of department, faculty coordinators and student club leads. It covers what I've built, what I've won, who I've mentored, what I can run for your students, and how to reach me. Contact, up front: pranjulrathour41@gmail.com · LinkedIn · GitHub · Blog on Dev.to · Bluesky What I've built Between February and July 2026, at Next Upgrad Web Solutions, I independently designed, built and shipped five production AI applications end to end, each wi...
Post a Comment
Read more