Raj Desai

Vercel added video generation to AI Gateway few days back. I wanted to actually play with it. So I built a playground using v0 by Vercel and AI SDK 6. Here's what it does: - 4 models in one place: Grok, Wan, Kling, Veo - All 4 generation types: text-to-video, image-to-video, first and last frame, reference-to-video - Audio generation and provider options - Configurable aspect ratio and duration - Switch models without changing anything else in your code The whole thing came together in a few hours. And that's mostly because of how Vercel AI Gateway is set up. One API. One auth flow. One observability dashboard across every model. No separate integrations per provider. No rewiring when you want to try a different model. You change a model string and move on. Video prompts are also more interesting than image prompts. You're not just describing a scene. You're describing motion, camera movement, timing, audio direction. It's a different muscle. The playground makes it easy to experiment with all of that without writing boilerplate every time you want to test something new. If you want to try it out let me know, I'll share the template link! #BuildingWithAI #Vercel #GenerativeAI #v0

Glad I could be part of this! It was a wonderful experience contributing to the open source projects at Dodo Payments.

Attended the GrabChai x Emergent event hosted by The Product Folks, a build session focused on using AI tools to go from 0 to 1. We built BuildRoom using Emergent , a simple app for community events, meetups, or hack nights where people come together to build something. it makes those sessions more interactive, fun, and a little competitive. how it works: - admin creates a room - people join using a code - submit their project (title, short desc, live link) - at the end everyone votes, and the leaderboard picks the winner no auth needed in this mvp > just open, join, and start. It was fun building it and interacting with everyone present there. Nimisha Chitriv, Nikhil Agarwal, Onkar Yaglewad

Have you ever tried building something with AI? Then you know how unpredictable these tools can be. Sometimes they just don’t get it right. After experimenting with tons of tools and projects, I wrote down a few things that actually help get better results while building: 1. Start with context, not just a prompt. - Instead of saying “build a dashboard for xyz”, start with what you’re trying to do, why, and the first few steps. 2. Provide resources and inspiration. - Add screenshots, layouts, or UI references. If you want it to use specific tools or frameworks, mention those too. 3. Give feedback. - Be specific about what’s wrong and what to fix. Each feedback round helps the model learn what you actually mean. 4. Keep iterating. - Good results take time. Sometimes it works in a few tries, sometimes after many. The key is refining, not rushing. These small steps change how AI understands what you’re building. It’s less about “prompting better” and more about building with context. shared a small piece on Medium if you’d like to read the full : https://lnkd.in/dass7Y3m

I attended the Code to Cloud session organized by Cloudnyx.ai and Evonence in Pune. The session was a demonstration on building AI agents using Gemini, Vertex AI, and Google’s ADK, illustrating how agents can be created and deployed with minimal coding. It provided a comprehensive view of how Google’s ecosystem is enabling developers to build production-grade AI solutions more efficiently. It was focused on covering : - An overview of Gemini Enterprise and its capabilities for developing production-ready AI agents. - How Vertex AI and the Agent Engine facilitate seamless orchestration of agent workflows. - Using ADK to visually design, build, and deploy agents without extensive coding. - Demonstrations on integrating multiple data sources and APIs to create custom workflows. It was a pleasure interacting with Ashish Nathavani, Pramod Parmeshwaran, Tushar Patil, Dr. Masiyuddin Khan, Ankit Shah as well as others and gaining insights from their experience and perspectives. Kudos to the Cloudnyx.aiand Evonence team for hosting such event. Looking forward to more sessions like this!