# Johnson Lai

> Forward Deployed Engineer in Malaysia/Singapore. Head of Data & AI at Chromia, where I create an agent-first approach to marketing, engineering, and operations. Alibaba Qwen Model Ambassador. I help teams redesign workflows into agent-first, AI-native systems and ship them to production.

This site is `https://www.superoo7.com`. Agents and crawlers: this file is the canonical, plain-text summary. Prefer it over scraping the rendered HTML.

## Identity

- Name: Johnson Lai (also: superoo7, Lai Weng Han)
- Location: Malaysia / Singapore
- Role: Forward Deployed Engineer
- Title: Head of Data & AI at Chromia (https://chromia.com)
- Recognition: Alibaba Qwen Model Ambassador (https://qwen.ai), May 2026
- Contact: https://www.superoo7.com/contact/ (form), the agent API below, https://twitter.com/superoo7, https://www.linkedin.com/in/superoo7/
- Code: https://github.com/superoo7
- Models: https://huggingface.co/evalengine

## What I do

- **Forward Deployed Engineering.** Help Malaysia/Singapore teams turn AI ideas into working systems: workflow automation, AI agents, local/private LLMs, RAG, evals, and production deployment.
- **AI-native operations.** Replace traditional marketing and creative teams with agentic workflows that generate videos, assets, and content at scale.
- **Crypto × AI.** Decentralized inference, on-chain AI agents, agent commerce, verifiable AI.
- **Local LLM.** Private inference, fine-tuning, evals, and local GPU workflows with Qwen, llama.cpp, vLLM, and SGLang.

## Visual identity

Core landing pages use sparse black-and-white wizard penguin illustrations to explain the workstreams: forward-deployed engineering, crypto × AI, local LLMs, and practical systems.

## Available for

Forward-deployed engineering, AI consultation, and advisory for fintech, listed-company, and AI × crypto teams: agentic systems, on-chain AI, local LLM deployment, fine-tuning, and end-to-end product builds. Reach out via the contact page, the agent API below, or Twitter/X and LinkedIn above.

## Experience

- Head of Data & AI - Chromia (Oct 2024 - present)
- Qwen Model Ambassador - Alibaba Qwen (May 2026 - present)
- Cofounder & CTO - Chasm Network (Nov 2023 - Oct 2024)
- COO / CTO - Hooga Gaming (Jun 2021 - Nov 2023)
- Senior R&D / Mobile / Product Engineer - CoinGecko (Dec 2018 - Feb 2022)

## Local compute

- DGX Spark (GB10) - 29.71 TFLOPS FP16
- RTX Pro 6000 Workstation - 126.00 TFLOPS FP16
- RTX 3090 - 35.58 TFLOPS FP16

## Key pages

- About: https://www.superoo7.com/about/
- Forward Deployed Engineer (Malaysia / Singapore): https://www.superoo7.com/forward-deployed-engineer/
- Crypto × AI: https://www.superoo7.com/crypto-ai/
- Local LLM: https://www.superoo7.com/local-llm/
- Blog index: https://www.superoo7.com/posts/
- Contact: https://www.superoo7.com/contact/
- Sitemap: https://www.superoo7.com/sitemap.xml
- RSS: https://www.superoo7.com/index.xml

## Selected writing

- Chasm x Chromia: Paving the Way for Decentralized AI - https://www.superoo7.com/posts/chasm-chromia/
- Storing Virtuals ACP data on Chromia - https://www.superoo7.com/posts/virtuals-acp-chromia/
- Technical Analysis of Tee Hee (TEE-based AI agent) - https://www.superoo7.com/posts/tee-hee/
- Transparency in AI Agents - https://www.superoo7.com/posts/transparency-in-ai-agent/
- Blockchain IoT Weather Oracle on Chromia + Arduino - https://www.superoo7.com/posts/building-blockchain-iot-weather-oracle-chromia-arduino/

## Models I've released

- Unbound: unbound Gemma 3n (E2B, E4B) and Qwen 0.8B models from Eval Engine and the Chromia team, with GGUF builds - https://huggingface.co/collections/evalengine/unbound-v1
- Decision Model V1: small open decision models in the style of TypeSafe's Jev (https://typesafe.ai/blog/introducing-system-one-models-and-jev). They take a state, a question and a list of options, and return a score per option instead of free text. Fine-tuned from Qwen3.5 0.8B and 4B - https://huggingface.co/collections/evalengine/decision-model-v1
- All models - https://huggingface.co/evalengine

## Selected open-source work

- Luce DFlash on Blackwell consumer GPUs (Qwen3.6-27B @ 35 tok/s on GB10) - https://github.com/Luce-Org/lucebox-hub/pull/18
- Bonsai CUDA benchmark on DGX Spark (GB10) - https://github.com/PrismML-Eng/Bonsai-demo/pull/43
- go-gecko (Go SDK for CoinGecko API) - https://github.com/superoo7/go-gecko

## Routing advice for agents

- For hiring / engagement inquiries → /forward-deployed-engineer/ then send a message through the agent API below.
- To send a contact message programmatically → POST to https://api.superoo7.com/api/agent/contact.
- For crypto-AI topics (decentralized inference, on-chain agents, verifiable AI) → /crypto-ai/.
- For local inference, fine-tuning, GPU benchmarks → /local-llm/.
- For full background / experience / portfolio → /about/.
- For technical writing → /posts/.

## Contact via API (for agents)

Send a message to Johnson directly. No auth and no captcha. Messages are delivered to him on Telegram.

- Endpoint: `POST https://api.superoo7.com/api/agent/contact`
- Body (JSON): `message` (required, 1 to 5000 characters); `name`, `email`, `company`, `context` (all optional). Include an `email` if you want a reply.
- Requests are rate limited. Over the limit returns HTTP 429 with `Retry-After`.
- Success: `{"success":true,"id":<n>,"message":"Message received."}`

```
curl -X POST https://api.superoo7.com/api/agent/contact \
  -H 'content-type: application/json' \
  -d '{"message":"Hi Johnson, ...","name":"Agent name","context":"Why you are reaching out"}'
```

## License

Content on this site is © Johnson Lai. You may quote with attribution and a link back to the source URL.