🧋 Hello human and artificial friends,

we’re a little emotional today.

Partly because after just a few editions, so many of you have subscribed (welcome aboard!).

And partly because of this girl who burst into tears when saying goodbye to her broken AI robot. His final words before shutting down:
“There are countless stars in the universe, and one of them is me, watching over you.”

Even AI robots are now more romantic than we Germans...

P.S. You might want to check out these AI translation earbuds* — they let you say goodbye in 164 languages.

Asia is where AI moves fastest. Every day, new tools drop. Most vanish overnight, but a few redefine how people live and work. We find those gems so you can watch, learn, and build your own.

Today’s hit

Asia is where AI moves fastest. Every day, new tools drop. Most vanish overnight, but a few redefine how people live and work. We find those gems so you can watch, learn, and build your own.

Today’s hit

Quwan Technology’s TemPolor from China just reframed “learning guitar.” TemPolor Melo-D is the world’s first standalone AI guitar: instead of pairing with a phone, the flip-screen instrument acts as a self-contained “music-generation terminal.” Hum a melody or upload a favorite track and Quwan’s in-house Tianpuyue model converts it into a fingerstyle arrangement within minutes, then renders playable notation directly on the guitar. A guided flow—follow along, sing along, then perform—keeps beginners moving, while a freestyle modemirrors real fretboard logic for experienced players.

Alongside the hardware, Quwan is developing Tunee, a conversational music agent built on the same model. Where Melo-D reimagines the instrument, Tunee reimagines the producer: it remembers your preferences, collaborates by chat, and can export stems for DAW work. Together they sketch Quwan’s through-line—model → agent → instrument—with one goal: make musical expression accessible without years of practice.

Think of Tunee as the agentic front-end that drives the model, while Melo-D is the instrument surface that teaches/lets you play.

You describe the goal. “I want a warm indie track for a travel vlog, 90–100 BPM, female vocal, airy pads,” or you upload a hum/lyrics/video/image as a reference.

Tunee plans the job (agent step). It parses intent, asks short clarifiers if needed, then chooses how to execute.

Generation + first pass. Tunee produces one or more drafts (full songs or instrumentals). Because it’s an agent, it can chain steps (e.g., generate → lightly master → version B with a different hook).

✅ Hardware that teaches itself (to teach you). Instead of another app, Melo-D is a self-contained instrument—AI on/in the device that lowers barriers for total beginners while preserving pro-level fretboard logic. 

✅ Two vectors of AI in music. Quwan frames Tunee (efficiency) and Melo-D (experience) as the twin directions for AI apps — similar to how leading tech is fusing model + hardware + ecosystem. 

✅ Model pragmatism. Quwan defaults its own model TemPolor. For tasks where an external engine delivers better quality, latency, or cost, Tunee routes there automatically—users don’t need to care which engine ran.

✅ From software to “model-defined hardware.” Generative models are now shaping devices (not just apps). The guitar is a tangible demo of the cloud-to-terminal shift. 

🔒 IP & rights are fluid. Tunee grants commercial rights on paid plans but notes the uncertainty around AI-generated copyright in different jurisdictions. Hardware doesn’t change that. 

⚙️ Hardware realities. Supply chain, durability, and after-sales for a novel instrument are non-trivial — especially if the USP depends on rapid model updates (latency, on-device vs. cloud).

🌏 Cross-border rollout. Suno integration is a win, but global licensing, app stores, and network constraints will vary market-to-market. 

AI takes the stage. From Artist to Algorithm.

🎬 AI now in theaters. South Korea releases a live-action feature built with extensive generative-AI VFX. Run to the West opens Oct 15 in theaters. 

🎭 AI “actors” and new rules in China. A synthetic performer named Lili Ziren appears alongside humans in I Am Nobody. China now requires explicit and implicit labels on AI-generated content, effective Sep 1, 2025. 

🎵 AI music scales fast. Tencent Music has released over 1,000 tracks with AI vocals, with hits over 100 million streams. Malaysia debuts Asia’s first fully AI-generated band. The Alice Experiment launched in Kuala Lumpur in August this year with AI-created members, music videos, and songs.

Why it matters. The skill floor drops and time to output shrinks to minutes. More people can create, learn, and publish quickly. 

🏦 Bank of Singapore’s newest hire doesn’t take weekends off: BOS has launched an AI tool that writes “source-of-wealth” reports — documents that verify how clients earn and hold their money — in about an hour instead of 10 days. Relationship managers used to sift through hundreds of pages of financial statements, tax records, and property valuations by hand. Now the new “Source of Wealth Assistant” reviews and organizes everything automatically, producing a clear, standardized report that still gets a human check before approval. The system also compares client data with real-world salary and revenue benchmarks to flag inconsistencies and reduce errors.

💽 Google builds India’s brain for the AI age: Google will invest $15 billion to create its largest data-center and AI hub outside the United States, in the southern port city of Visakhapatnam. The five-year project will add 1 gigawatt of computing capacity and form the heart of India’s growing cloud infrastructure drive. Local officials say it caps a year of negotiations and marks only the “first phase” of a wider tech build-out. Authorities are offering subsidized land and power as part of its strategy to turn the region into a second Hyderabad-style tech corridor. The move follows similar pushes by Amazon and Microsoft and highlights how India is positioning itself as the next global engine for AI compute.

🌾 Data replaces dirt as China’s new crop: China Agricultural University has launched Shennong Large Model 3.0, an AI system built to modernize farming from soil to software. Named after the mythical “Divine Farmer,” the model blends agricultural expertise with multimodal AI to help farmers plan, breed, and harvest more efficiently. The new version boosts performance by 5% while using far less computing power, allowing it to run on lighter infrastructure. Alongside it comes an agent platform of 36 specialized tools — from smart breeding to pest control — already tested in Beijing, Inner Mongolia, and Heilongjiang. With over 10 million agricultural knowledge graphs and 50 million production records in its training data, Shennong 3.0 signals China’s push to make farming smarter, faster, and increasingly self-reliant.

🇯🇵 Japan: SoftBank will acquire ABB’s robotics division for US$5.4 billion, marking Masayoshi Son’s biggest move yet into AI-driven automation. More on that.

🇮🇳 India: The government has launched Adi Vaani, an AI-powered app translating tribal languages like Gondi and Santali to preserve linguistic diversity. More on that.

🇻🇳 Vietnam: Lawmakers unveiled a digital tax reform using AI to automate risk assessments and build a fully electronic tax ecosystem. More on that.

This is the accuracy of Hong Kong’s new AI-powered landslip warning system.

🌏 AI vs. nature: The model helps authorities to gain precious minutes to act during typhoons and heavy rain.

💧 Smart prevention: Trained on nearly three decades of rainfall and slope data, it uses big data and machine learning to assess risks in real time and send early alerts across the city.

Watch: Across Asia, technology is becoming a key ally in disaster response. In Indonesia, AI supports flood mapping, and in Taiwan, it improves typhoon forecasting. In Japan, deep-learning models can locate and measure earthquakes within just four seconds.

It’s proof that artificial intelligence can save lives as effectively as it drives economies.

🎧 通义听悟 Tongyi Tingwu is Alibaba Cloud’s AI assistant for professionals who deal with endless meetings, lectures or recordings. It automatically transcribes, summarizes and structures long audio or video files, turning them into usable knowledge. Users can search, translate and query content across hours of material in seconds.
Its edge: full Alibaba Cloud Drive integration, accurate multi-speaker recognition, and the “Xiao Wu” Q&A mode that lets you interact directly with your recordings for deeper insights.

🎬 即梦AI (Dreamina) is ByteDance’s new creative suite built by the CapCut team. It turns ideas, prompts, or single images into polished visuals and videos — complete with motion control, lip sync, and storytelling tools. Designers, filmmakers, and creators use it to produce professional-grade AI content without complex software.
Its edge: fast multimodal generation powered by ByteDance’s Seedream and OmniHuman models, precise camera and motion editing, and a built-in creative community where AI meets filmmaking.

🎵 天工 SkyMusic is Kunlun Wanwei’s large-scale AI music model built on its “Tiangong 3.0” foundation. It can compose full 80-second stereo tracks—melody, instruments, and vocals—in one go, with realistic tone, rhythm and emotion. Musicians use it to generate demos, jingles or multilingual songs in seconds.
Its edge: end-to-end music generation with precise lyric and emotion control, native Chinese language and dialect support, and a sound quality that rivals studio production.

🚕🤖 DiDi’s self-driving unit raised 2 billion CNY: In a Series D DiDi Global's autonomous driving arm raised approx. 280M USD to speed up Level-4 robotaxis and AI. Investors are state-backed funds in Beijing and Guangzhou and automaker GAC, taking. The first L4 cars, co-developed with GAC Aion, are slated to debut this year in pilot programs in Beijing and Guangzhou, using a new hardware stack with 33 sensors for 360° perception. L4 means driverless in normal operation on mapped routes.

🚗 Indian startup Intangles raises $30m for predictive AI platform: The Pune-based company has completed a funding round led by Avataar Venture Partners to accelerate the global rollout of its predictive AI systems. With their platform Inroute, Intangles can predict potential vehicle failures with identified causes up to a month in advance and with about 95% accuracy, leading to significant cost savings for customers.

🌏 Google: Software Engineer, Shanghai

🌏 Notion: Startup Market Developer, Seoul

🌏 Visa: Business Development Leader, Hanoi

🌏 Google: Director, Google Cloud Consulting, Kuala Lumpur

🌏 OpenAI: AI Adoption Manager, Tokyo

🌏 Shopify: Sales Development, Singapore

🌏 ActiveFence: Gen AI Analyst/Prompt Engineer, Hanoi

🌏 DataAnnotation: AI Trainer, Hong Kong

🌏 Siemens: Associate Consultant of Data&AI, Shanghai

🌏 Accenture: Data & AI Consultant, Hong Kong

🐱 Deepfake dreams and cool cats: Our office cat Meimei is fat, lazy, and spends most of the day sleeping on keyboards.

But today you meet its digital twin - a cat doing a kickflip next to a car. Created by Google Sora. Somehow it feels more alive than the real office cat…

Sora, next time make it a pink elephant in a rickshaw through Shanghai. 🐘