Full Case Study Coming Soon
AtpDeco
The Multi-Surface Architecture That Carried a Marketplace Into a B2B Platform
I designed the multi-surface architecture beneath AptDeco's marketplace, building two AI engines (Predictive Logistics Engine, Data Capturing Engine), one Recommend-Not-Command Trust Contract, and a four-tier authorization model across three user surfaces. Drove ~70% operational efficiency gain, ~25% CS workload reduction, and carried the business to scale. Expanded from Tri-State to nationwide and into its first enterprise white-label deal within the 18-month investor window.
0→1
AI Trust
MULTI-SURFACE
Predictive Systems
Expert Encoding
Summary
The investor's goal was one enterprise B2B deal in 18 months. Three projects moved up the backlog to reach it:
→
Internal Capacity Management Tool
→
Overhaul Marketplace Redesign
→
Internal Warehouse Management Tool
From the UX perspective, I introduced an asymmetric-risk question. "What happens if the marketplace succeeds but the operator layer cannot absorb the demand that success creates?"
Executive alignment by the end of the meeting. The capacity platform shipped first as the foundation everything else would run on.
The expert's brain was the system
The Head of Ops ran logistics off pattern recognition and spreadsheets. Years of accumulated judgment that couldn't be captured through traditional interviews. So I sat with her as she worked. She narrated her decisions live, examined her own habits in real time, and engineering identified the data dependencies the interface was exposing. The handoff wasn't specs with implementation notes. It was a map of semantic intent.
One trust contract ran through two engines
While engineering wanted maximum model authority, Ops wanted absolute human control. The resolution reframed the question from "how much should the model decide" to "how does the model earn the right to decide more."
Recommend-not-command became the contract. It ran through two engines.
→
The Predictive Logistics Engine reconciles three parties on capacity and pick-up & delivery scheduling.
→
The Data Capturing Engine pulls transactional history to recommend listings, suggest pricing, and pre-fill details.
Same principle, two engines, different problems.
The architecture earned its keep as leverage
The same Data Capturing Engine that helped sellers price their listings also
→
powered cross-buyer recommendations
→
calculated delivery-batching windows
→
ompiled market intelligence on brand resale velocities
One engine, four jobs. AptDeco used that market intelligence to approach manufacturers with data rather than a sales pitch, and signed its first white-label enterprise deal inside the 18-month investor window. The marketplace built the foundation. The B2B business consumed it.
Built
→
Multi-Surface Architecture
→
Recommend-Not-Command Trust Contract
→
Predictive Logistics Engine
→
Data Capturing Engine
→
Tiered Authorization Model
→
Expert-Encoding Method
