Build the gate. Check the claim.

Decision-Grade AI is a one-day, in-person buildathon during SF Tech Week, hosted by Ezio at Industrious Palo Alto with infrastructure partners HarnessRouter, TokenRouter, and Keenable, on October 10, 2026.

Hear from researchers and practitioners, find your team, then spend 105 minutes on one of three challenges about putting AI decisions to the test.

No AI safety background required. No coding needed for Before You Act. We’ll provide the tools, mentors, and challenge materials—the full specs are revealed at kickoff.

APPROVAL GATE — TECHNICAL BUILD

A trust score doesn’t make a decision. Your wrapper does.

An AI draft says “send it.” The score looks clean. But the discount is wrong—and nobody checked. That’s the moment this track is about. Build a decision layer that reads the signals, cross-checks a second source of truth, and calls APPROVE, HOLD_FOR_REVIEW, or BLOCK before the action goes out—with a reason you can defend.

Choose a high-stakes workflow—a support lead sending a customer email, a paralegal submitting a memo, an analyst publishing a client note—and decide what evidence is good enough to let it run. A green score isn’t proof the facts were checked.

You’ll build against a local simulated Ezio API, so no live key or paid credits needed. Everything runs on your own laptop—the starter kit is shared at kickoff, and the final judging set is released mid-build.

Repo: https://github.com/ezioai/Approval-Gate
Setup Guide: https://drive.google.com/file/d/1wXWrNrK9NSl2GC-sSyPbgDJJryc346sJ/view?usp=sharing

Passphrase : 4KdJ_Ezio_SF26

Submit here →

BEFORE YOU ACT — NON-TECHNICAL WORKFLOW CHALLENGE

An AI gives you a confident answer. Should you act on it? Pick a small everyday decision—a refurbished laptop to buy, a software plan to pick, a trip to plan—and write down the facts that would actually change your answer.

Craft a prompt, save the exact answer, then validate it with Ezio: inspect the sources, weigh what’s supported, contradicted, or unanswered, and build a one-page decision workflow—who acts on what, by when, including the branch where your key question never gets answered.

No code, installation, or APIs required—a checklist, table, flowchart, or readable paper page works. Submit one PDF with your workflow and complete Ezio report. Materials provided at kickoff.

Repo: https://github.com/ezioai/Before-You-Act

Submit here →

BUILD A TRUST AGENT — TECHNICAL BUILD

An LLM gives an answer. A verification tool checks it and produces a trust report. Then what? Build an agent that takes a real Ezio trust report as input and decides what happens next—approve, rewrite, escalate, block, or pull in another source—using the evidence, not just the top-line verdict.

Generate one of the four sample reports on the Ezio Try page, decide who’s relying on the answer and for what, and show your agent make the call. Any stack, any harness—but your agent must work on any of the four reports, not just the one you picked. Submit a 1–2 minute demo video.

Repo: https://github.com/ezioai/Build-Trust-Agent

Submit here →

PUT EZIO TO THE TEST — SHARE FEEDBACK, ENTER THE RAFFLE

Open to eligible attendees across all tracks, including podium winners and attendees not entering a judged track. Try live Ezio with event-provided access on a non-sensitive use case. A live challenge run already counts—no extra exercise required.

Click Feedback in Ezio’s top menu to share feedback, then Enter the raffle → We’ll draw 5 winners of $50 Amazon gift cards at random during wrap-up. Positive, critical, or mixed feedback has equal odds and does not affect project scores or the judged Best Use of Ezio award. Keep confidential and personal information out.

THE TOOLKIT

• Live Ezio: event-provided product access for the challenges and the optional feedback activity, using non-sensitive inputs. ezioai.io
• TokenRouter: $50 in API credits per participant, redeemed through the event-day participant portal. Optional. tokenrouter.com
• Keenable: $1,000 in web-search credits per participant—independent crawler, index, and ranking for AI agents, over REST or MCP. Signup code provided at kickoff. Optional. keenable.ai · docs.keenable.ai
• Mentors: available during the build window.
  Please join HarnessRouter Discord Server for the Hackathon Event -- https://discord.com/invite/KZGNyZmDW

PRIZES & PERKS

💵 HarnessRouter Cash Prize: $500 cash
⭐ Best Use of Ezio: 1 month of free unlimited Ezio access

🏆 1st Place: $1,000 in TokenRouter API credits
🥈 2nd Place: $500 in TokenRouter API credits
🥉 3rd Place: $300 in TokenRouter API credits

Free for every hacker: $1,000 in Keenable web-search API credits, $50 in TokenRouter API credits, and Ezio Beta access.

$2,300+ in announced prizes, plus $1,050 in developer credits for every participant — and 5 × $50 Amazon gift cards in the feedback raffle.

SCHEDULE — ALL TIMES PACIFIC

10:30 AM — Check-in, coffee, and team mixing
11:00 AM — Technical kickoff and demos
11:40 AM — Opening talk: Kiana Jafari, Stanford Center for AI Safety
12:00–1:00 PM — Expert panel: OpenAI, Stanford, Microsoft Research, and IBM
1:00–1:15 PM — Challenge reveal and team formation
1:15–3:00 PM — Build window: 105 minutes for ALL tracks, with mentor support; submissions due at 3:00
3:00–3:20 PM — Track scoring: judges select the top project per challenge
3:20–3:45 PM — Judges’ panel awards 1st/2nd/3rd + Best Use of Ezio
3:45–4:00 PM — Wrap-up + raffle draw

FEATURED VOICES

• Kiana Jafari — Executive Director, Stanford Center for AI Safety
• Bryant McCombs — GTM, OpenAI
• Clark Barrett — Computer Scientist, Stanford
• Wesley Deng — Senior Researcher, Microsoft Research
• Sarath Swaminathan — Research Scientist, IBM

WHO SHOULD COME

Engineers, students, researchers, designers, founders, and anyone who wants to test AI against evidence. Come solo or with a team. Teams have 1–4 members and choose one judged track.

Venue: Industrious Palo Alto, 1881 Page Mill Road, Palo Alto, CA.

RSVP on Partiful for venue entry. Every competing team member must also register on Devpost.

Event details: hack.ezioai.io
Questions: [email protected]

• Accuracy
  Information on the main page!