Hi, I'm Weiren Lan

I’m Weiren Lan, an AI architect and applied AI engineer based in Taipei, Taiwan. For 8+ years I’ve built production AI across speech, audio, NLP, and computer vision, from training models to shipping them to millions of users. Lately most of my work is LLM agents and agentic coding.

This blog is where I write down what I learn along the way.

Want to work together or just talk AI? Reach me on LinkedIn.

At a Glance

• 2M+ users served by audio AI products I built (noise eraser, meeting-ink)
• 15+ production ML/LLM models shipped across audio, NLP, and CV
• 90% lower cost, 9x faster transcription after I rebuilt the pipeline
• 7 days: the time it took to ship ByeType, an iOS AI voice keyboard, with Claude Code
• 3 granted patents in audio processing and equalizer tuning

Featured Project: ByeType

ByeType is an AI voice keyboard for iPhone (with a macOS companion). You talk the way you normally talk (“um… three, no wait, four”) and it writes the clean sentence.

• Real-time speech recognition, then an LLM rewrites the text to fit the app you’re typing in
• Fixes technical terms, lets you edit with voice commands, and supports custom style prompts
• Runs fully on-device with WhisperKit, or with cloud providers (OpenAI, Anthropic, Gemini, ElevenLabs) using your own API key
• Supports 20+ languages

How I built it: I’m a Python and AI algorithms person, not a Swift engineer. I built ByeType in a 7-day sprint over the 2026 Lunar New Year holiday, using Claude Code for about US$330 in total usage. The workflow was plan → review → implement → UI/UX polish → test → CI/CD to App Store Connect. Two things I learned: agentic coding still leaves gaps in feature state management, and deep audio knowledge still matters a lot.

Read the full write-up: Building an AI Voice Keyboard App in Swift Using Claude Code.

Experience

Founding AI Engineer → Lead AI Algorithm Engineer · DeepWave Intelligence

Taipei · Jul 2020 – Jun 2026

I joined DeepWave as its first AI engineer when the team had fewer than four people, and I built and led the AI team for six years.

• Speech and audio AI: noise-reduction and source-separation models deployed to 2M+ users and 7,000+ hours of audio (noise eraser); singing-voice separation and voice conversion models
• Meeting intelligence: an enterprise meeting system with 20 summary templates in 8 languages, plus a bilingual transcription and LLM terminology-correction pipeline for 6 languages (90% lower cost, 9x faster)
• Real-time agents: multimodal voice agents (Yamaha AI Assistant) on Gemini 2.0 Flash and GPT-4o, using tool calling and Plan–Reflect–Action loops
• LLM workflows: content generation for summaries, PRDs, and educational material, including a PRD generator built on Claude Skills
• Enterprise knowledge: planned a PoC to capture expert tacit knowledge, combining expert interviews (CDM/ACTA), transcription, and grounded LLM extraction
• Leadership: set the AI technical direction, owned dataset strategy (10,000+ samples, YOLO at 98% accuracy), and mentored junior engineers

Software Engineer, AI/ML · UnlimiterHear

Taipei · Jul 2018 – May 2020

My first job, at a company started by an assistive-technology foundation to build hearing-assistance algorithms. I brought deep learning to an acoustics team, built a speaker recognition system, and deployed a real-time recognition API on AWS.

Open Source

Projects I built that people found useful:

I also maintain awesome-llama-resources (★ 44), a curated list of Llama resources.

Social Innovation: InternLens

Before AI took over my life, I co-founded InternLens (實習透視鏡), a volunteer project that made internships in Taiwan more transparent.

It started in 2016, when I was a grad student and saw friends working unpaid “internships” that were really just jobs. I listed every internship at a campus job fair in a spreadsheet: 66 of 108 had no pay listed or were unpaid. After I posted the numbers, the organizer had companies fill in the missing details, and the hiring platform added a “paid” badge.

Next I launched an anonymous survey on internship pay and conditions. It was shared 3,000+ times in two days, and a small team formed around it. Together we built:

• A Facebook community of 9,200+ members
• A website with 459 real internship reviews, later handed over to GoodJob
• An infographic that a legislator cited during a question session in Taiwan’s Legislative Yuan, plus coverage in Business Today and Storm Media

Some organizations changed their intern pay and policies as a result. I wrote about what we learned (in Chinese): Social innovation projects, you can start one too.

Don’t ask why nobody is doing this. You are the “nobody”. (g0v)

Education & Recognition

• M.S., Bio-Industry Communication and Development, National Taiwan University. Published deep learning research on ultrasound imaging in Computer Methods and Programs in Biomedicine
• B.S., Electronic Engineering, National Yang Ming Chiao Tung University
• 3 granted patents in audio processing and equalizer adjustment
• Semi-finalist, 2024 Taiwan Llama Competition (top 8 of 30 teams)
• Hugging Face Audio Course (2023) · AI Accelerators Certification, Turing Certificates (2022)

Elsewhere

• LinkedIn: linkedin.com/in/weiren-lan, where I post about speech AI, agentic engineering, and startup lessons
• GitHub: github.com/MIBlue119
• Medium: @willylan, older notes on hearing tech and AI research
• ByeType: byetype.com