Delphi · Synthetic populations as a computational substrate · Google I/O Hackathon 2026

Delphi is a new computational primitive — synthetic populations as a service. Built solo in one day on Gemini 3.5 Flash for the Google I/O Hackathon, May 23, 2026.

You ask any question — e.g. "Will the Fed cut rates in Q3 2026?" — and hundreds of Gemini 3.5 Flash sub-agents, each role-playing a different American persona generated from US Census–aligned demographic axes and each grounded in live web, reason about it in parallel. In ~60 seconds you get back a probabilistic forecast with confidence interval, the strongest reasons for and against drawn from agents' own reasoning, where demographic groups diverge most, a striking outlier quote — all synthesized into a Wall Street Journal–style summary by a final Gemini call.

Live-validated on stage at N=500 with a +66.2 percentage point shock response (13.3% → 79.5% when surprise May CPI printed 2.1%).

It's not a chatbot. It's not a copilot. It's a new primitive.

Code (MIT): https://github.com/SankarSubbayya/delphi

Built with: Gemini 3.5 Flash · FastAPI + asyncio · Next.js + React + Three.js (R3F) · Tailwind · Tavily (search grounding) · uv

Validation shipped:
- 27 / 27 automated tests
- Adversarial persona stability — 4.92 / 5 in-character, 0% drift to centrist mean
- Cross-model drift — Gemini 3.5 Flash vs 2.5 Flash showed a 87.5 pp per-agent success-rate gap on identical prompts
- End-to-end stress test at N = 200 (207 s wall-clock, 89% per-agent success)

Sankar Subbayya · Shack15, San Francisco · 2026-05-23

Видео Delphi · Synthetic populations as a computational substrate · Google I/O Hackathon 2026 канала Sankaranarayanan Subbayya