Suggestions
Chengyin Liu
Chengyin Liu is the Co-founder and Chief Technology Officer (CTO) of Slab, a software company focused on creating tools for team collaboration and knowledge management. He has played a pivotal role in scaling Slab to achieve multi-million dollar annual recurring revenue (ARR) and profitability, supported by notable investors such as Matrix Partners and CRV.1
Professional Background::
- Slab: Chengyin co-founded Slab in June 2017 and served as CTO until December 2023. Under his leadership, the company developed innovative features, including a product that transforms GitHub activities into human-readable changelogs using large language models (LLMs) and a popular browser extension for web typography inspection, which has over 3 million weekly users.12
- Previous Experience: Before Slab, he was involved in various software development projects, including building a dashboard for Wandoujia, an Android management tool. He also designed an embeddable store that gained significant traction upon launch.1
Education:: Chengyin Liu holds a Bachelor of Science degree in Computer Science from the University of Illinois at Urbana-Champaign.3
Current Focus:: At Slab, Chengyin emphasizes creating intuitive software that enhances teams' ability to store and organize information efficiently, which he believes is crucial for effective execution.4
Highlights
🤠 Excited to share my latest fun project: AI Changelog (https://aichangelog.com)! It crafts changelogs for any repo and frees you from sifting through walls of commits. You can use it to keep teammates and users updated or to upgrade dependencies with confidence.
🤖 LLM Model: I started with gpt-3.5-turbo, moved on to claude-3-haiku, and am now using gpt-4o-mini. Watching the progress in this short period of time has been mind-blowing - my prompting struggles vanished with 4o-mini, and it’s blazing fast. We’re truly in a new era of possibilities.
🛟 One game-changer: OpenAI’s structured output. Tools like Instructor often needed retries, and Claude occasionally struggles with JSON syntax. We have superpowers with these models, but there’s still work to be done in that last-mile delivery.
😓 Not-so-fun fact: gemini-1.5’s massive context window seemed promising, but frequent infinite output loops have made it impractical for now.
💕 Thank you for reading! Check it out, and I'd love to hear your thoughts!
