I'm Bill Zhang. By day I build LLM agents at Pinterest. By night I build software you can talk to, which is how I ended up with 36 hackathon wins. This page is one of those builds. Go on, ask it something.
Latency, interrupted sentences, tool calls that half-fail, a model that answers the wrong question with total confidence. At Scale AI I shipped a multi-agent system that works through up to 20K denied claims a week, and owned the eval framework that moved audit accuracy from 64% to 88%. Almost all of that came out of the failure cases.
A 911 console that triages live calls, a whole bank that runs inside a phone call, a food-rescue network that dispatches by voice, game tape you can question mid-replay. Several of them need nothing but a phone. No app, no internet.
At a hackathon I pick one idea and make it actually work. Dispatch AI was a phone line, a fine-tuned triage model, and a dispatcher screen. It took the $50K grand prize at Berkeley against 900+ people.
Most people can speak several times faster than they type, with a fraction of the effort. And while you talk, your hands are free for whatever else you were doing. That is why this site listens instead of making you click, and why TalkTuahBank handles balances, transfers, and bill pay by voice alone. No app, no data plan.
★ 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.
Every one of these was a weekend: build it, demo it, win over the judges. All during college. $150k+ in prizes.
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 information, one page, no talking required. Read it below or take the PDF; both match what the rest of this site says.
“Send me your resume.” → opens the download.
Ask about a project and the agent looks it up, describes it, and opens it on the page. Retell carries the audio, Deepgram hears you, GPT-5.6 Terra thinks, ElevenLabs talks back. A Python backend runs the agent and tells the browser where to scroll.
Browser mic → Retell client SDK over WebRTC. Deepgram transcribes the stream as you speak.
Retell hands the transcript to a FastAPI backend, where a GPT-5.6 Terra agent (OpenAI Agents SDK) runs with this site's content in context, nine registered tools, and a jailbreak guardrail in front.
Agent emits structured tool calls: display_project(id), display_resume_page(), search_projects(query, ...), and six more.
navigation.py turns each display_* call into a navigation metadata event; the browser converts it to a bus command. 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 reply streams back through Retell into an ElevenLabs voice while the scroll is still finishing, so the answer and the page land together.
// server: navigation.py turns a display_* tool call into
// { type: "navigation", page: "project", project_id: "dispatchai" }
// browser: metadata event → bus command → scroll
client.on("metadata", ({ metadata }) => {
const action = metaToNavigationAction(metadata);
if (!action) return;
VoiceBus.emit(action.command);
requestAnimationFrame(() => scrollToSection(action.scrollTo));
});
// every section subscribes to the same bus
VoiceBus.on((cmd) => {
if (cmd.type === "open") setOpenId(cmd.id);
if (cmd.type === "scroll") scrollToSection(cmd.id);
});