~/about $ cat about.md
Johnson Lai
Forward Deployed Engineer in Malaysia and Singapore. Head of Data & AI at Chromia, Qwen Model Ambassador.
Head of Data & AI at Chromia, Qwen Model Ambassador
I鈥檓 a Forward Deployed Engineer in Malaysia and Singapore. I sit with a team, learn how the work actually gets done, and replace the slow parts with AI systems that run in production. Not a demo, not a slide deck.
I鈥檓 self-taught and a UWA alum. Today I run data and AI at 
Chromia, where I鈥檓 building an agent-first way of running marketing, engineering and operations. I鈥檓 also a Model Ambassador for 
Qwen. Before that I was cofounder and CTO at Chasm, COO and CTO at Hooga Gaming, and a senior R&D and mobile engineer at CoinGecko.
What I do
Most of my time goes to three things.
-
Forward Deployed Engineer I join teams in Malaysia and Singapore, redesign one workflow at a time, and ship the agent system that runs it.
-
Crypto 脳 AI Decentralized inference, on-chain agents, agent commerce, and AI whose outputs you can check.
-
Local LLM Running, fine-tuning and evaluating open models on my own GPUs with Qwen, llama.cpp, vLLM and SGLang.
Work with me
I take on forward-deployed engagements where AI meets crypto: agent systems, on-chain AI, local LLM deployment, fine-tuning, and full product builds.
Send me a message, or find me on X and LinkedIn.
Experience
| Role | Company | Period |
|---|---|---|
| Head of Data & AI | Chromia | Oct 2024 to now |
| Qwen Model Ambassador | Alibaba Qwen | May 2026 to now |
| Cofounder & CTO | Chasm Network | Nov 2023 to Oct 2024 |
| COO & CTO | Hooga Gaming | Jun 2021 to Nov 2023 |
| Head of Product, Advisor | Pacer | 2022 |
| Co-organizer | ETHKL | 2020 |
| Senior R&D, Mobile and Product Engineer | CoinGecko | Dec 2018 to Feb 2022 |
| Fullstack Engineer | PhytoMark | Jun 2017 to Jan 2018 |
Selected work
Models
- Unbound (2026). The first unbound models from Eval Engine and the Chromia team: Gemma 3n E2B and E4B, plus a Qwen 0.8B variant. GGUF builds are included for llama.cpp.
- Decision Model V1 (2026). Small open models in the style of TypeSafe鈥檚 Jev. Instead of writing text, they take a state, a question and a list of options, and score each option. Fine-tuned from Qwen3.5 0.8B and 4B.
- Everything else I publish is on Hugging Face.
Open source
- DFlash for Blackwell consumer GPUs in Luce (2026). Adds DFlash support for the 5090 and GB10. Qwen3.6-27B runs at 35 tok/s on a GB10, about 3脳 faster than vLLM with DFlash and 9脳 faster than vLLM at bf16, with OpenAI-compatible tool calling. (announcement)
- Bonsai CUDA benchmark on DGX Spark (2026). Community numbers for Bonsai 8B, 4B and 1.7B on a GB10 with 128GB unified memory and CUDA 13.0.
- go-gecko. A Go SDK for the CoinGecko API.
- steem-web-wallet. A web wallet for the Steem blockchain.
- UnicornAutoStaker. An auto-staking bot for Hooga.
Products
- Tatarot. Tarot readings generated by AI.
- Weave. A drag-and-drop AI workflow builder in the spirit of n8n and Flowise, built at Chasm.
- Pacer Wellness App
- AlphaOnChain
- Bull & Bear Merch
- CoinGecko Widgets and the CoinGecko mobile app
Writing
Talks
Hackathons
- ETHKL 2023. Finalist with PenguMon.
- ETHGlobal Tokyo 2023. PPSwap won Quantstamp鈥檚 Most Creative Solution, a Worldcoin honorable mention and a Polygon pool prize.
- Utopian Hackathon 2018. Steem Web Wallet.