Creator of Sveltepress · 537 stars
Dongsheng Zhao
AI Full-Stack Engineer · 9 years
I put LLMs inside production systems — regulatory reports drafted by Claude and OpenAI, a multi-chain risk engine underneath, a public API around it and the workbench on top. Design, build, test, deploy: on my own.
Open to full-time remote or contract · UTC+8 · Hefei, China · working English
- 9
- years shipping
- 723
- GitHub stars
- 80
- public repos
- 4
- open-source orgs
What I am building now
LLMs inside a crypto compliance platform
A crypto AML/KYT compliance SaaS. I own the AI layer and the whole stack beneath it — screening engine, public API, workbench, and the pipeline that puts it on a server.
Regulatory writing, done by the model
On-chain risk data and rule hits are fed to Claude and OpenAI, which draft STR/SAR reports, compliance policies and rule-set proposals — replacing a step that used to be written by hand.
KYA / KYT screening engine
Multi-hop fund-path tracing that surfaces high-risk counterparties, sanctions and blacklist hits, then returns a graded risk verdict.
Configurable rule sets
Path-level detection routed by scenario — deposit, withdrawal, CDD, ongoing monitoring — so compliance teams change policy without shipping code.
Public API, v2 and v3
Async polling with API-key auth, address-monitoring alerts, webhooks and quota billing, across Ethereum, Tron and Base.
The workbench
Next.js 16 and React 19 in TypeScript: screening, batch checks, graph exploration, reports and monitoring, on SWR and Zustand.
Ship and keep it up
Led a full cloud migration, built the CI/CD on GitHub Actions — atomic Go binary swaps, layered backups, rollback in seconds — and own production incidents.
Rewrote path investigation
Uploading the whole graph became server-side detection with gzipped parallel paging. A single screening run got dramatically faster.
Open source · AI and agents
Tools that drive coding agents
I spend a lot of time inside coding agents, so I build the parts that are missing.
agent-feishu-channel
49Bridges Claude Code and Codex sessions into a Feishu (Lark) bot — session lifecycle, plus two-way relay of messages and tool calls. Drive a coding agent from the chat client you already have open.
codsh
6A terminal coding agent on the DeepSeek harness, with its own tool-call loop and interaction design. One sentence to /ship, and it comes back verified.
deskfolk
6Alpha. A local-first macOS app for persistent AI teammates — bots that chat one to one, join groups and hand work to each other, over a split-pane workbench with real terminals and a flow board per job. Bring your own OpenAI-compatible endpoints and MCP servers.
Open source · Svelte ecosystem
Sveltepress, and what grew around it
Maintained since August 2022. Most of my GitHub stars come from here.
Sveltepress
537A content-first site builder on SvelteKit: SSR, SSG and CSR; Svelte 5 components written straight into Markdown; versioning, i18n and Pagefind search; themeable, with a default theme and a fully typed API end to end.
svelte5plus-calendar
5A full calendar with zero dependencies — month, week, day, year and agenda views, drag and drop, recurring events, i18n and dark mode.
svelte-json-discovery
5The discovery.js JSON struct view, lifted out into a standalone Svelte 5 component you can drop anywhere.
# Getting startedWrite Svelte right here:<script> let count = $state(0)</script><button onclick={() => count++}> {count}</button>Getting started
Write Svelte right here:
really clickablePlayground
Built because they were fun
Some of these have more stars than the serious ones. I have made peace with it.
v-rap
27A Vue 3 library whose entire purpose is embedding the Vue author's rap video in your app.
joueur
27A tiny, playful music player.
siyuan-plugin-*-repl
14Live Vue, Svelte and React REPLs embedded inside SiYuan notes. Three plugins, one idea.
reditor
8An online resume builder with decent templates. Yes, I wrote a resume tool before I wrote this page.
Nine years, six domains
Where the work has actually been
Regulated finance, hospital equipment, consumer audio, lending at scale, internal platforms. Different constraints, same job: make it correct and make it ship.
- Current
Crypto compliance and on-chain risk
LLM-generated regulatory output, multi-hop tracing, configurable rule sets, a public API and the cloud it runs on.
- Yiyi Data
Internal developer platforms
Planned and led an in-house low-code platform — core engine plus its own domain design language. Delivery on internal projects got about 30% faster, and duplicate work went away.
- GE Healthcare China
Medical imaging
Technical management and core feature work on the x20 system and the China AW imaging platform.
- Bilibili · MissEvan
Consumer audio at scale
Product features and frontend iteration, plus the internal component and utility libraries the team built on.
- Oufei Tech, Beijing
Overseas lending and merchant platforms
Lending H5 flows, back-office and a data aggregation platform behind a loan book over $100M and a million-plus merchants. Build time down 50%, dev start and HMR twice as fast, Lighthouse 90+ in production.
- AsiaInfo · Uinnova
Telecom-grade frontend
Stack selection, project scaffolding, core pages and CI; built the shared libraries the department kept using after I left.
The stack I reach for
- AI
- Claude · OpenAI · DeepSeek · structured output · tool-call loops · agent CLIs
- Languages
- TypeScript · Go · Rust · Java · Python
- Frontend
- React 19 / Next.js · Svelte 5 / SvelteKit · Vue 3 · Vite / Rollup / Nx · Tauri · D3 / ECharts
- Backend
- Node.js · Go · PostgreSQL / MySQL · Redis · REST and async jobs
- Infra
- AWS · Tencent Cloud · Docker · GitHub Actions / Jenkins · Linux · systemd deploys with rollback
- Chains
- EVM and Tron on-chain data · Ethereum · Tron · Base
Contact
Tell me what you are building
If you need someone who can take an AI feature from an idea to a thing on a server, without a handoff in the middle — write to me.
Open to full-time remote or contract
Based UTC+8. A full European working day overlaps; so does US East in my evening.
Email dongshengzhao47@gmail.com