San Francisco

Hi, I'm Michael.

I'm a product manager at Pallet, where we build AI agents for companies that make and move things. Before that: the Army, Berkeley, JPMorgan, and Wells Fargo.

Work

I grew up building while working summers at my family's construction company. The itch to build never left, the materials just changed.

01 — Granite
enterprise AI · regulated finance
live

An interactive architecture for building AI in spaces with high reputational risk and federal regulations — where a wrong answer is worse than no answer.

Trace a request through intent → entitlements → orchestration → grounded response → LLM-as-a-judge → audit, and watch it refuse rather than guess.

Along the way, it shows where choosing deterministic over generative pays off. Click into each node to see the thinking behind it.

view project →
Granite — the architecture tracing a request
02 — Strata
forward deployed case · messy in, structured out
live

Strata, sediment pressed into ordered layers, is one extraction engine, not a chatbot, for the paperwork problems industrial companies have carried since their founding: messy data, no in-house engineering.

Where Granite is maximum governance, this is maximum simplicity: it reads the mess, writes the structured record, and asks a human only when it isn't sure.

Built on real data: interviews with two operators with 35+ years inside a national installer, plus the three summers I spent working in its branches.

view project →
Strata — switching from scheduling intake to acquisition integration
03 — Cairn
side project · sierra nevada trip planning · live data
live

A cairn is a pile of rocks stacked to mark the route — now mostly used as a landmark or for an Instagram photo. This one tells you whether you can still get a permit.

I built it for my friends and me after 5+ years of planning Sierra trips the manual way: one site for the latest beta, another for weather, and, most annoying of all, refreshing for permits.

Describe a trip in plain language and Cairn returns routes with live permit availability, a packing list tuned to current conditions, and a GPX file for Garmin & Strava.

view project →
Cairn — typing a trip request and opening a matched route
04 — Commute
side project · bay area transit · live data
live

My friends and I couldn't rely on the live times Apple Maps showed for our Golden Gate Transit and Presidio GO commutes. So I built a tracker for each, connected straight to the source: the agency's own real-time feed.

Every arrival time on screen says where it came from: the bus's GPS, the agency's prediction, or just the timetable. When a trip is canceled or the feed goes stale, the page says so instead of guessing.

It also knows the rules the feeds leave out, like Presidio GO's pass-only rush-hour runs and its weekend schedule on federal holidays. Tap the bus you're on and it counts down your stops, then tells you when to pull the cord.

About

I enlisted in the Army at 18 and spent four years as an intelligence analyst — briefing generals in an underground bunker near the Korean border, driving tanks across the Mojave.

Then Berkeley: Political Economy and Finance (Haas BASE Program, 2021), splitting time between reading geopolitical theory and reconciling balance sheets.

Post-Berkeley I joined JPMorgan as an associate product manager, rotating across three teams — onboarding, an innovation team hunting the bank's next bet, and the platform behind the digital bank. That's where I figured out I'm a 0-to-1 PM. I'd rather ship scrappy and hand off at scale.

At Wells Fargo I was the first AI product manager in the wholesale digital organization. I recruited a senior architect and an engineer, and we delivered a governed GenAI experience for the bank's FX clients. When adoption stalled on data migrations rather than the AI, I turned the work into a platform strategy and secured funding to build it out.

Now I'm at Pallet, where we build AI agents that run day-to-day operations for manufacturers, distributors, and logistics companies.

michael.cerp@gmail.com LinkedIn ↗

Photos

Off the clock, usually at altitude

Summit ridge of Mt. Shasta
Mt. Shasta · 14,179 ft · 2024-05-11 · 41.4092°N 122.1949°W
Mt. Whitney
Mt. Whitney · 14,505 ft · 2024-10-04 · 36.5785°N 118.2923°W
Crater of Pico de Orizaba
Pico de Orizaba · 18,491 ft · 2025-02-16 · 19.0303°N 97.2683°W
Palisades Tahoe
Palisades Tahoe · 9,006 ft · 2026-04-03 · 39.1981°N 120.2873°W