Join ambitious builders, stay ahead of AI, build your own agents, create opportunities for your family, and turn ideas into real products.
SiliconRoshi Membership isn't a subscription to AI classes. It's membership in an AI Builder ecosystem — where you Learn, Build, Connect, and Create Opportunities to become a Super Individual in the AI era.
SiliconRoshi brings together people who are ambitious, decisive, curious, and willing to take action — not just people who want to 'learn AI,' but people who actually help each other move things forward.
Help each other become Super Individuals in the AI era.
Members' children may have opportunities to apply for internships with WukongDojo — our AI startup — and get exposure to real AI startup projects. Help prepare the next generation for the AI era.
Note: Internship placements are opportunity-based and go through application and matching. Preference is given to member families where appropriate — no placement is guaranteed.
AI is evolving too quickly for anyone to follow alone. Members learn, filter, and judge together what's actually important.
Stay ahead of AI — together.
Don't stop at 'I learned it.' Build a real Agent on WukongDojo.ai (or your own platform) and put it to work on a real problem.
Don't just learn about agents. Build one. Launch it. See where it can go.
The people you meet here are not just classmates. They may become collaborators, customers, mentors, talent, partners, project teammates, referral sources, and future co-founders. You can also grow from member into a Project Lead / Program Co-sponsor and bring others along.
The person you meet here today may become the person you build something important with tomorrow.
Membership isn't the ceiling. It's the middle rung of the Academy's official cert ladder. If you want to, you can keep climbing.
Publish an agent · pass the L1 quiz → first certificate.
Become a member to enter the track · attend the course · ship a real AI app (3+ real users) · human review.
↑ Members enter hereCo-sponsor at least one Academy-approved event that records revenue → lead Academy projects and events.
L3 · AI Principal is this Academy's program-lead tier — the formal path for Project Leads & Program Co-sponsors.
Joining as a Member buys you a standing seat: every Tuesday 5:30 PM PT, Bill takes you through it personally — either building a digital employee that actually works, or making one key concept click. One hands-on, one mind-on; together they're the real skill: "see a piece of work, know how to hand it to AI." The 30-minute sprint is mine, not yours — I build it in front of you, you take a full week to make it your own, and we review everyone's work the following week. Topics follow the AI frontier and what you ask for · always updating, never graduating. Below are the directions in the pipeline — tap any for detail, and requests are welcome.
Starting from an Idea — Build and Deploy live. Get a real website online globally in 30 minutes, at zero cost, accessible from anywhere — your one-person company's first employee: a storefront that never clocks out.
The 30-minute sprint is mine, not yours: I build it in front of you, you take a full week to make it your own. By the end of session one you understand: in the AI era, the distance from Idea to Product is completely different.
The real tool stack Bill uses every day — the one that lets siliconroshi.ai run as a one-person operation. Eight weapons · one job each · no bloat:
Wispr Flow (voice input) · Claude Work (multi-agent workbench) · Claude Code (AI coding) · Nano Banana Pro (image generation) · ChatGPT (general Q&A) · DeepSeek (Chinese-first reasoning) · Gemini NotebookLM (turn your own files into a study partner) · Gemini in Gmail (AI where your inbox already lives).
Not a list of tool names — the lesson is: which weapon for which situation. (Sourced from S3.5 "AI 时代极速前进" · Wukong's Arsenal chapter.)
A whole session dedicated to AI-driven PPT — from idea to finished deck.
Research, structure, content, imagery, motion, presentation pace — end-to-end, live. Stop spending a whole evening making slides by hand.
Enter real AI Coding. From "can't write code" to being able to Build real software and products with Claude Code.
Idea → Code → Debug → Build → Deploy end-to-end, one person completing the loop. AI Coding isn't just a programmer's tool — it's becoming a core Builder skill.
Claude Work is Anthropic's multi-agent workbench — complex workflows, multi-turn tasks, agentic collaboration.
A dedicated session on how it differs from Claude Code, when to reach for Work vs Code, and how to compose a workflow that runs itself.
The Agent Loop is the heart of every agent system — plan → tool call → observe → reflect → next step.
Without understanding the loop, you only write prompts. With it, you design agents that drive themselves forward. This session covers ReAct, reflection, planning, and error-recovery loop patterns from first principles to real usage.
💡 The buzzword Loop Engineering you keep hearing? This is it. Don't let the new label fool you — it's just designing this agent loop.
MCP (Model Context Protocol) is Anthropic's open standard — letting agents connect to any data source, tool, or service via one protocol.
Writing an MCP server = turning your private data / tools into agent capabilities. This session walks you from zero to a working MCP server — protocol shape, transport, authorization, and a tour of ecosystem servers already available.
ChatGPT knows a lot — but it doesn't know your company, your project, your files, your knowledge.
RAG (Retrieval-Augmented Generation) is the first-layer solution to making AI truly understand your own content. Principles, embeddings, chunking, re-ranking — a real RAG pipeline built live.
Organize your docs, notes, and SOPs into an LLM-readable structured wiki.
Not a Notion-style human-first wiki — a wiki purpose-built for AI consumption, letting agents instantly look up, cite, and traverse links to reason.
Ontology is the skeleton of your world — concepts, relations, constraints — lifting AI from "reads text" to "understands the domain."
Without ontology, agents guess. With it, they reason like domain experts. This session covers building an ontology for your business from scratch.
Weave wiki + ontology into a Knowledge Graph — your own knowledge network.
Agents traverse graph edges for multi-hop reasoning, answering complex "across-doc, across-concept" questions. Build, query, maintain — one session end-to-end.
💡 Some now call this Graph Engineering — this is the session. Build knowledge into a graph · let agents reason across it · same thing, new label.
Context Engineering is the hottest term right now — but it's not magic. It answers one question: given a limited context window, what exactly do you feed the model.
Feed too much and the model gets distracted, burns money, and gets dumber (context rot). Feed too little and it has no idea what you mean. S08 RAG, S09 LLM Wiki, S10 Ontology, S11 Knowledge Graph all cover "how to store knowledge well" · this session covers "once it's stored, which piece to pull on each call, how to order it, how much to include" — retrieval strategy, context compression, layered memory, prompt-assembly order, and why "stuff everything into the prompt" is always wrong.
After this session you'll see it clearly: Prompt Engineering is writing one good sentence · Context Engineering is managing the whole context. The latter is the watershed that takes an agent from demo to production.
Skills are the lightest way to give an agent new capabilities — each Skill is a reusable capability module.
This session unpacks Skill principles, authoring, and composition — with a live custom Skill for your own domain.
Plugins are how agents reach the outside world — calling APIs, querying databases, driving external services.
Skill is "agent's internal capability"; Plugin is "agent's external reach." When to use which, how to design a plugin interface — covered end-to-end.
Hooks are the mechanism for adding interception points to an agent workflow — inject custom logic before, after, or on-error for a tool call.
Hooks are the most-overlooked yet highest-leverage primitive for agent stability. This session covers hook lifecycle, common patterns, and production-grade best practices.
Harness = run your agent in a testable sandbox · Eval Loop = a repeatable evaluation that keeps telling you how good it actually is.
Without a harness you only have vibe checks. With one, you can systematically move an agent from 60 to 90. This session covers test-set design, pass-rate vs pass-score, regression triggers, and the real-user-data → eval-set feedback loop.
💡 The trending Harness Engineering is exactly this session — a testable sandbox + eval loop for your agent. New word, old craft.
GStack is an open-source AI stack toolkit from Garry Tan (YC president) — infrastructure builders use to spin up agents and AI products fast.
Bill walks you through installing it, running it, hacking on it, and turning it into a daily weapon for your own startup or business. Open-source = zero license fees, full control, hack anything you want.
GBrain is the "brain" layer in Garry Tan's open-source ecosystem — weaving knowledge, context, and reasoning into an actual "brain" your agent can rely on.
Paired with GStack: Stack provides tools, Brain provides thinking. What you end up building isn't a scatter of isolated prompts — it's a real, working AI System that accumulates experience over time.
Make your agent actually collect money. This session walks end-to-end through Stripe Payment Links, Checkout Sessions, Webhook + idempotent confirmation, subscriptions vs one-time, invoices, refunds, and tax basics.
siliconroshi.ai's own payment closed-loop (the exact system that sold you this membership) is dissected live — dual-source webhook + lander idempotency, UNIQUE(stripe_sid) backstop, audit-table design. One session to understand what "billing in the AI era" looks like.
Security = your agent can't be broken by prompt injection, data leak, or tool abuse · Governance = who on the team can change prompts, add tools, read logs, and audit changes.
Required prerequisites for the Enterprise tier: secrets management, least-privilege, rate limiting, abuse detection, rollback plans, and a SOC-friendly audit trail.
Once you've built an AI System that runs, we teach you to make the UI look professional.
CALM = Clear · Accessible · Legible · Minimal. Less is more · every element earns its place · one idea per screen. Avoid the AI-era "slop UI" — cluttered, gimmicky, broken layouts. After this session, what you build doesn't look like a hobby project at first glance.
Scale everything above to enterprise level — team collaboration, permissions, audit, compliance, data governance, cost control.
Building agents solo and deploying agents company-wide are two different games. This session covers the pitfalls and patterns for enterprise AI deployment — for CTOs, tech leads, and founders planning to sell AI into enterprises.
The topics above teach you to build AI. This track teaches you to turn AI into a business.
Upcoming weekly live sessions will cover: SEO (traditional search engine optimization · Google/Bing/Baidu) · GEO (Generative Engine Optimization · get your content cited by ChatGPT / Claude / Perplexity) · Blogging (the AI-era content engine · a workflow that ships ten posts a day) · Marketing (AI-powered growth, ads, email, social) · Sales (cold outreach, CRM, pipeline, objection handling — all AI-powered) · One-Person Company (OPC) (running ops, finance, and support through an agent stack — no hires needed) · Startup 0 to 1 (from idea to MVP to first paying customers — Bill's own startup playbook).
Charter + Founding members: all included, no additional fee.
Strong tech, wrong direction — pure waste. This session starts from finding the direction: how to use ChatGPT to systematically surface ten money-making niches, how to rank them, and how to filter out the "looks hot but pays nothing" traps.
Not a pep talk — Bill runs the workflow live: exact prompts, how to make ChatGPT estimate TAM, price point, competitive density, and entry cost, ending with a scored candidate list. After this session, you can find the direction worth focusing on for a year — in one evening.
Content · product · marketing — every one of them starts with finding what's actually being searched and discussed. Wrong topic, and even flawless execution lands in cold water.
This session walks through the full trending-topic radar toolkit: Google Trends · X (Twitter) real-time trending · Reddit hot / rising · YouTube trending · TikTok Discover · Xiaohongshu · Douyin hot list · Product Hunt · Hacker News · Perplexity Discover — plus how to build an AI agent that aggregates every signal into one auto-refreshing hot-topic dashboard for you.
In the AI era, content is leverage — and camera is the doorway. A lot of tech-first builders stall here: great product, nobody knows who they are.
This session covers going on camera: beating camera fear · minimum viable lighting + audio (a phone, a mic, one lamp is enough) · pacing on camera · eye contact · posture · designing a 30-second hook · format differences across Douyin / Xiaohongshu / YouTube Shorts / X video · and building a sustainable weekly on-camera cadence. Bill dissects his own on-camera mistakes live.
Building it and having nobody use it — the most painful trench every builder falls into. This session walks through the real path from 0 to 10 to 100 users — no paid ads, only the right hand-crafted acquisition moves.
Covers: Cold DM / Email that doesn't read as spam · how to surface yourself compliantly inside Reddit / Xiaohongshu / WeChat communities / X replies · how to find the first truly paying user (nothing like the first 9 free ones) · how to turn early users into a referral source · how to use AI agents to automate outreach and follow-up for your first 100 users.
The truth: finding your first user is harder than shipping the product — and every builder has to do both by hand at least once.
💬 Students are welcome to raise questions and suggest topics — we adjust the schedule based on what you actually want to learn.
What you see is the current 20+ topics in the pipeline · more will be added over time. Course schedule and content are subject to actual scheduling · exact titles, order, and pace may shift as the course evolves.
Academic background
Undergrad at Fudan University, master's at Tsinghua University — where he stayed on as faculty — then Ph.D. at National University of Singapore, followed by postdoctoral research at University of Pennsylvania.
During Ph.D. research, collaborated with computational linguist Martha Palmer to propose what became known as Wu-Palmer Semantic Similarity. The paper is cited thousands of times, and Wu-Palmer Similarity is implemented in mainstream NLP tools and systems still in use today.
Industry experience
Moved from AI academic research into industry with 20+ years of Silicon Valley software engineering and management experience. From frontline Software Engineer all the way to Software Development Director — led large engineering teams, owned Enterprise Software, shipped Production Releases, did Product, did Engineering Management, and holds multiple U.S. and Chinese patents.
What Bill does today
Back on the AI frontline: personally builds AI products, personally builds Agents, personally builds a Startup. What SiliconRoshi AI Academy teaches is not "where's the button" or "how to use this tool" — it's decades of Research + Engineering + Product + Management + Startup re-applied to today's AI era: from Idea to Architecture, to Product, to Deploy, to entering the real world and creating value.
The proof: www.siliconroshi.ai and www.wukongdojo.ai — two real, running AI systems Bill led the team to Build and Deploy · already have paying users. Follow the Academy, and students Build the same kind of thing.
SiliconRoshi AI Academy's Tuesday live sessions are primarily led and curated by Bill. Occasional guest sessions from invited Builders / Practitioners will complement the curriculum — Bill remains the Academy's curator and lead teacher.
For: professionals · founders · college students · high-schoolers · anyone who wants to truly Build with AI
Apply AI to your current work · lift your output · build products · build a startup · run a one-person company.
College Research · School Selection · Personal Story · Essay · Activities · Resume · Interview Prep — the student's own story, expressed better with AI.
Build Projects · Portfolio · Resume · Research Opportunities · Interview Prep · real project opportunities via WukongDojo. What employers want to see today is what you've built.
Job Research · Resume · Portfolio · Networking · Interview Prep · Technical Prep — the goal is not to spam 500 resumes; it's to become genuinely more competitive in the AI era.
WukongDojo — wukongdojo.ai — is the AI Agent platform Bill leads. Members build their own Agents on it and put what they learn to work on real problems the same week.
Not another 'community of like-minded people' cliché — a place that actually helps you move what you're working on forward.
Inside the member community, these questions get asked — and answered. The person you meet here today may become the person you build something important with tomorrow.