I used to think the biggest AI challenge was choosing the best model. Turns out, it's choosing th...

I used to think the biggest AI challenge was choosing the best model. Turns out, it's choosing th...I used to think the biggest AI challenge was choosing the best model. Turns out, it's choosing th...

The network for creativity

Join 1.25M professional creatives like you

Connect with clients, get discovered, and run your business 100% commission-free

Creatives on Contra have earned over $150M and we are just getting started

Back to feedPost

Md Samsuzzoha

• Jul 28

I used to think the biggest AI challenge was choosing the best model.

Turns out, it's choosing the right model for the right job.

One engineering team rebuilt their AI infrastructure by routing simple tasks to self-hosted open-weight models and reserving premium APIs only for complex reasoning. The results were impressive: 82% lower AI costs, 41% faster responses, and 3.6× higher throughput.

The biggest lesson for me wasn't about replacing commercial AI. It was about treating AI like infrastructure, not a single API.

Smart routing, workload-specific models, continuous monitoring, and flexible architecture can reduce costs without hurting user experience.

As a full-stack JavaScript developer, I find this approach far more practical than chasing leaderboard rankings. Great engineering is about balancing performance, scalability, privacy, and cost.

Detail at https://blog.webdevlab.org/chinese-open-source-ai-models-cut-ai-costs.

Would you build an AI-first product with one model, or a routing system that picks the best model for every request?

ai llm genai

blog.webdevlab.org

Chinese Open Source AI Models Cut Our AI Costs by 82%

Discover how an AI-first SaaS company cut AI infrastructure costs by 82% using Chinese open-source AI models, hybrid inference, and self-hosted deployment.

Chadd Weston

max

• Jul 25

bento boxes for AI client done in Framer

UI Design UX Design Framer Web Design ai

Asif Aleem

pro

• Jul 25

Bento reads best when the tiles vary in weight so the eye gets an anchor before it hits the dense cells; on a data-heavy dashboard like this, giving the pipeline card clear priority over the smaller metric tiles is what keeps it from feeling like a wall of equal boxes.

Hossain Jahed

• Jul 25

Stop settling for barebones UI. 🛑

If you’re building an AI startup, a generic chat input won't cut it. You need a component that actually feels intuitive, responsive, and production-ready.

Stop rebuilding the wheel and start using the input designed for the AI era. 🚀 Link below 👇

AI Development Web Development Google Gemini React Tailwind CSS buildinginpublic webdev uicomponents UI Design

Hossain Jahed

• Jul 25

get it for free in my component library https://easemize.21st.dev/components/ai-chat-input

easemize.21st.dev

Ai chat input | Community Components | 21st

A modular, single-file React input bar designed for generative AI interfaces and agentic chat workloads, built on the shadcn/ui and Tailwind CSS architecture. Features dynamic spring-physics transitions (cubic-bezier engine) for height scaling, state-morphing typography, an integ

Mmenyene Tom

• Jul 24

Here is submission for the Envato × Contra Human by Design challenge.

I set out to answer a simple question: What happens when AI becomes part of an animator's creative process instead of replacing it?

I started out with Envato 3D assets, I used Envato's AI tools to explore ideas, generate concepts, test transitions, and build each scene one decision at a time. Every shot was refined through countless iterations until the visuals, sound, and pacing came together as a single story.

For me, this project isn't about what AI created.

It's about the creative choices that turned those generations into something meaningful.

Created using Envato's Gen AI and Stock Assets Library, Edited and composited using PremierPro and Directed by My creativity.

envatochallenge premierepro AI Art 3D Animation Adobe Premiere Pro Envato

Effiong David

• Jul 24

This is one of my best Submission because the human impact to the final result is Undeniable. When I watched the results I felt it was one video, but having to go through the video that explained the process of how you brought it all together is amazing. BRAVO!