I make AI feel like a real product.
Most AI demos crumble under a real conversation. I obsess over the seams: latency, interruption handling, tool-use, fallback. The work that survives contact with users is the work I want to do.
--66b75db74f9f0264 x-nextjs-stale-time: 4294967294 x-nextjs-prerender: 1 x-next-cache-tags: _N_T_/layout,_N_T_/page,_N_T_/ vary: RSC, Next-Router-State-Tree, Next-Router-Prefetch, Next-Router-Segment-Prefetch
I'm Bill Zhang, Applied AI Engineer at Scale, voice-first builder, serial hackathon winner. This site listens. Talk to it, scroll it, drag it around. It'll rearrange.
Most AI demos crumble under a real conversation. I obsess over the seams: latency, interruption handling, tool-use, fallback. The work that survives contact with users is the work I want to do.
Phones. Smart glasses. Browser sidebars. Pinned terminals. Wherever the interface is least visual, I'm trying to make it more useful: for banking, education, public safety, food rescue, sports tape.
Thirty-six hackathon wins is a lot of weekends. The discipline of 36 hours sharpens the instinct: pick the smallest demo that proves the hypothesis, build the spine first, decorate last.
Most people on Earth still type slower than they speak, and a lot of them never learned to type at all. Every product I touch tries to bend the floor of access a little closer to the ground.
★ Grand Prize, UC Berkeley AI Hackathon 2024/dispatchai↗ deep diveEmpathetic 911 dispatcher co-pilot. Aggregates emergency calls, triages severity, extracts context, and keeps humans in control.
★ Google Challenge, LA Hacks 2024/adapted↗ deep diveLectures that talk back. A voice tutor with Gemini-driven slides and emotion-aware pacing.
★ Grand + Goldman Sachs, HackUTD 2024/talktuahbank↗ deep diveVoice banking over a regular phone call. No internet, smartphone, or banking app required.
★ MLH GitHub, CruzHacks 2023/slugloop↗ deep diveReal-time UCSC loop bus tracker. Helped students see live campus bus locations and reduce pressure on metro routes.
★ NLX Overall, AI ATL 2024/soundsearch↗ deep divePhone-call voice guidance for complex websites. A caller gets step-by-step spoken help while the page highlights what to do next.
★ 2nd Place, Hack Dearborn 2024/swarmaid↗ deep diveDecentralized food rescue agents. Matches surplus suppliers to food banks and shelters with real-time logistics planning.
Hackathon culture is a forge. You start with nothing, you ship something, you defend it on a stage. These are selected public wins from a broader run with $150k+ in prizes.
Enterprise Generative AI Platform team. Building agentic automation, LLM evaluation workflows, and multi-agent systems for enterprise AI.
NowShipped LLM evaluations and multi-agent assistants for customer-support workflows after starting as a senior AI intern building LLM-powered assistant prototypes.
Built an emergency-response AI platform with Berkeley SkyDeck funding. Solo-engineered the voice AI and telephony integration behind the dispatcher demo.
Co-founded a 4K-member hackathon community with sponsor backing, helping student builders find teams, ship faster, and learn from winning projects.
Founded and shipped the UCSC real-time bus tracker, turning live campus transit data into a student-facing map for loop buses.
Taught Python with a challenge-based curriculum, helping new programmers build fundamentals through hands-on coding problems.
Engineered a real-time device dashboard and cross-platform backend for connected hardware workflows.
Coached USACO students from Bronze toward Silver mastery through algorithm practice, debugging, and contest prep.
Mainly competed during college and still jumps into select events. Most of the projects on this page started as weekend builds.
Click a card to flip it.

Graduate-level ML, applied AI, agentic systems. Built voice-first products throughout.

Finished a 4-year program in 2.5 years. Caught the hackathon bug.

South Bay. First wrote code here.
Same story, traditional format. Click through below or grab the PDF; both stay in lock-step with everything else on this page.
“Send me your resume.” → opens the download.
Most portfolios are read-only documents. This one is an agent. You speak, and the structured tool calls coming back from the LLM rearrange the page in real time: filter projects, scroll, expand a deep-dive, jump to a specific section. The voice agent is the navigation.
Browser mic stream → Retell client SDK. Low-latency WebRTC pipe to a hosted agent.
Retell agent runs an instruction-tuned model with this site's content + a small tool registry as its world.
Agent emits structured tool calls: filter_projects(tag), focus_project(id), open_project(id), scroll_to(section).
A pub-sub on window broadcasts each tool call. No prop drilling, no router. Sections opt in.
Each section subscribes to the events it cares about and animates its own state. The site rearranges itself.
The agent confirms in spoken English while the page settles. The user keeps both hands free.
// every section subscribes to the same bus
VoiceBus.on((cmd) => {
if (cmd.type === "filter") setFilter(cmd.tag);
if (cmd.type === "focus") setFocusId(cmd.id);
if (cmd.type === "open") setOpenId(cmd.id);
if (cmd.type === "scroll") scrollToSection(cmd.id);
});
// the agent's tool calls become events
agent.onToolCall("filter_projects", ({ tag }) =>
VoiceBus.emit({ type: "filter", tag })
);