Devon Sun

Founder at AI Builders @theaibuilders | IRL AI Events in 🇸🇬 Singapore | 🇯🇵 Tokyo | 🇺🇸 San Francisco Bay Area

[AI Builders @theaibuilders]

300+ builders, 40+ projects, 1 Day. Agent Forge Hackathon in Silicon Valley is on 🔥

A full day of builders, teams, prototypes, demos, and live AI agent projects — all in one room.

Thank you to every participant who brought the energy, curiosity, and ambition. Huge thanks to our partners & sponsors: @agentfield_ai @ActionbookHQ @nosana_ai @qwen_cloud @qoder_ai_ide @zeaburapp @Zai_org @palebluedot_ai @Butterbase_ai @evermind @UofBeta @llama_venture

Watch the recap and relive the vibe 📷
#AgentForge #AIHackathon #AIAgents

[AI Builders @theaibuilders]

Wanna know what builders are working on in an AI hackathon? 😆

Watch this 👉

Together with @agentfield_ai @ActionbookHQ @nosana_ai @qwen_cloud @qoder_ai_ide @zeaburapp @Zai_org @palebluedot_ai @Butterbase_ai @evermind @UofBeta @llama_venture

#aibuilders #AgentForge #AIHackathon #AIAgents

[AI Builders @theaibuilders]

300+ builders at Agent Forge Hackathon: Build OpenClaw AI Assistant -- a full-day, in-person hackathon and shipped AI assistants with live data access, autonomous database management, and cloud deployment. 🚀

Key takeaways:
- Giving teams production-grade tools from the start -- serverless databases, web scraping APIs, deployment platforms, long-term memory -- changes what's possible in a single day.
- The best projects came from teams that pressure-tested their assistants against real-world workflows, not just polished presentations.

Huge thanks to @agnesai_sapiens @PingCAP @zeaburapp Bright Data- for powering every team's stack. 🥳

#AIBuilders #AgentForge #OpenClaw #AIHackathon #Singapore

[Nosana @nosana_ai]

Builders getting hands-on with Nosana in Shanghai 🇨🇳

Together with @theaibuilders, a workshop explored how developers can build and deploy AI agents using Nosana’s decentralized GPU network.

Always great seeing people learning, experimenting, and pushing ideas forward 💚

[AI Builders @theaibuilders]

Incredible energy at Beta University @UofBeta for "AI Coding with Qoder"! 🚀 @qoder_ai_ide

We explored why 2026 is the breakout year for AI-native coding.
The shift is real:
Code Completion (30% gains) 👉 Collaborative Coding (30-100% gains) 👉 Autonomous Programming (100%+ gains)

Big thanks to our experts from Microsoft, Apple, and Momen for the insights. The AI-Native era is here. 🦾

#AICoding #AIBuilders #Qoder #SiliconValley #AIAgents #SoftwareDevelopment

[Devon Sun @devonsun_ml]

I’m incredibly excited for the Agent Forge Hackathon together with @YaokaiJiang and @XinYu146992 from Momen @Momen_HQ this Sunday! The energy from the community has been overwhelming—we are completely oversubscribed 🙌

This response is a true testament to Singapore’s hunger for innovation and the thriving ecosystem of AI builders we have here. It’s inspiring to see so many people ready to get hands-on and push the boundaries of what’s possible.

I can’t wait to see everyone on Sunday as we collaborate with Momen to build, deploy, and scale the next generation of AI Agents. Let’s turn ideas into production-ready apps!

See you all there! 🚀
https://t.co/LgAAm68FPQ

#AI #Hackathon #Singapore #Momen #AIAgents #BuildInPublic #TechCommunity #AgenticAI

[AI Builders @theaibuilders]

Just wrapped up our meetup in Tokyo, and the energy was something else! 🇯🇵

We spent the night diving deep into @qoder_ai_ide, and honestly, the reaction from the Tokyo community was incredible. It wasn't just polite interest—local engineers were genuinely glued to the demos.

The shift from just coding to true agentic building is happening, and Tokyo is all in. Thanks to everyone who joined us—can’t wait to see what you build next!

#AIBuilders #Qoder #TokyoTech #DevCommunity #AgenticAI

[Devon Sun @devonsun_ml]

AI agents are receiving a user interface upgrade with MCP Apps. 🚀

I have developed a real-time Bitcoin Price Dashboard that integrates directly within an AI chat interface using MCP Apps. This eliminates the clutter of text and presents clean, interactive charts.

For those interested, here is the step-by-step tutorial:

https://t.co/TJbww3I1JV

#AIBuilders #MCPApps #AIAgents

[Kuu🍣 @Fumiya_Kume]

Qoderの公式Meetupが東京上陸🎉
Cursor と Kiro の良いところを融合させた感じのAI IDE、リポジトリのWIkiをローカルで作成したり Jetbrains の IDEに対応してるのが特徴的な部分か。
他は、各種AI IDEに搭載されてる機能はある程度網羅してて良さそう、値段も安いし。

参加するとQoderクレジットがもらえるっぽい。
#PR

[Devon Sun @devonsun_ml]

Zeabur @zeaburapp just launched AI Hub - the unified API gateway that changed how I build AI applications 🚀

Instead of juggling multiple API keys for Claude, GPT, Gemini, DeepSeek, and others, I use ONE API key for everything. OpenAI-compatible, so zero learning curve.

I tested 7 AI providers side-by-side, and you can check the full comparison report in the article:
- Alibaba: qwen-3-32
- Anthropic: claude-haiku-4-5
- DeepSeek: deepseek-v3.2-exp
- Meta: llama-3.3-70b
- Moonshot AI: kimi-k2-thinking
- OpenAI: gpt-4o-mini
- xAI: grok-4-fast-non-reasoning

Perfect for developers and AI Builders who want to experiment with different models without the integration headache.

What's the best AI model for your application? 🤔
https://t.co/OtqL0eYVIU

#AIModels #DeveloperTools #AIBuilders

[Nosana @nosana_ai]

BUIDL Day in Singapore is around the corner! 🌏

🤝 Join us for a day of connection, learning, and building — with keynotes, panels, and a mini hackathon featuring GPU credits & prizes.

Brought to you by Nosana and @theaibuilders, with special participation from @Sogni_Protocol, @ogpucommunity, and @mastra.

⚠️ Spots are limited — save yours today: https://t.co/GAYTSqbQ30

[Devon Sun @devonsun_ml]

Want to use the latest Seedream 4 image editing model but tired of clunky and expensive AI image tools?

I wrote a quick guide on how to spin up your own private AI image studio using Lobechat on Zeabur with Fal API. You'll be using the state-of-the-art, super powerful Seedream 4 model to do cool edits, like placing logos into photos, in just a few minutes.

Full guide inside!

https://t.co/ikFGEeUcvi

#AI #AIBuidlers #GenerativeAI #ImageEditing #Seedream4 #Zeabur #LobeChat #falapi

[Devon Sun @devonsun_ml]

🧵 4/4 step 4: Zeabur manages DevOps, builds in 3-5 mins—app live! 🎉 focus on creativity, not deployment.
P.S. grab your FREE $5 credits here: https://t.co/We0hs1uxoD

[Devon Sun @devonsun_ml]

🧵 1/4 hi vibe coders! 😄 community feedback: tools like V0, Bolt, Lovable deploy easily, but Cursor, Windsurf, VS Code Copilot feel tricky for newbies. 🤔 Zeabur extension bridges this—sharing 3-step walkthrough for one-click deploy from Cursor!

step 1: build app in Cursor. prompt: "build simple Next.js todo app." it creates app with TypeScript, Tailwind, localStorage. run `npm run dev` at localhost

[Devon Sun @devonsun_ml]

🧵 3/4 step 3: AI scans: detects Next.js, TypeScript, Tailwind, React, localStorage—no configs needed. create project, choose region

[jason @jxnlco]

I published a brain dump of all my thoughts on contact engine engineering so far

What else should I read about?

https://t.co/XmxO2Rk5uX

[Google Labs @GoogleLabs]

Still "vibe testing" your LLMs? 🤔 Introducing Stax, an experimental developer tool to streamline LLM evaluation with custom and pre-built autoraters.

Go try it now! https://t.co/8602kt0RVy

[Devon Sun @devonsun_ml]

During the development of our own AI SaaS product, we quickly realized that the quality of the LLM's output was directly tied to the quality of the context the model has access to. This led us to develop a crucial metric internally we call Context-Problem Fit (CPF).

Context-Problem Fit (CPF) measures how well the provided context equips an LLM to solve your problem. We formalize it with this heuristic index:

CPF ≈ UPR / LTR = (UX × Precision × Relevance) / (Latency × Token Cost × Risk)

*You can remember them as "Upper" for UPR, "Later" for LTR because we want the numerators up, and "don't let it drag later" for denominators 🤣

Where:
- UX: user experience of the LLM-based applications
- Precision: noise minimized, fact checked
- Relevance: semantic + task alignment
- Latency: time to complete the task
- Token Cost: total input + output tokens consumed
- Risk: hallucination / compliance / safety exposure

Great agents fail more from poor CPF than weak models. Always evaluate: Does this context boost the probability of a correct, useful answer? If not, refine via retrieval, compression, or filtering.

By focusing on CPF, we've been able to build more sophisticated and dependable AI features than we ever could through prompting engineering alone.

You can learn more about Context-Problem Fit (CPF) form this article we wrote:
https://t.co/aVew6Iviyp

[Nosana @nosana_ai]

Want to run Stable Diffusion on decentralized GPUs? We got you!

Here’s a step-by-step guide showing how to deploy it on Nosana, from setup to generating your first images 🖼️

Check it out: https://t.co/UKidyVdkfO