Holiday Pay Tracker
An AI-assisted operational tracking tool designed to improve visibility, consistency and accountability around holiday-pay review workflows.
How We Built It • Phase 4
Applied AI &
Work Projects
Phase 4 moves the portfolio beyond describing AI capability and starts documenting AI-assisted tools being used to solve real operational problems. The goal is to show the work, the thinking behind it and the practical value—without exposing company-sensitive information.
PHASE 4 • OPENStarted Sep 23, 2026First milestone: launch the AI Work Projects portfolio and document three operational tools with public-safe project evidence.
Milestone 01
AI Work Projects goes live
A new AI Work Projects section connects the website’s AI story directly to professional operations work. Instead of presenting AI as a concept, it documents tools designed around real workflow needs and explains the problem, solution, operational use and lessons behind each project.
View AI Work Projects →Inbound Report
An AI-assisted reporting workflow that organizes inbound activity into a clearer multi-site operating view and reduces repetitive manual reporting work.
DC Trailer Audit Tool
An AI-assisted audit workflow designed to create a more structured, traceable record of trailer and outbound quality checks.
Public portfolio standard
Show the real work without exposing internal information
The project pages use real screenshots from the working tools, but the public versions are sanitized. Company-sensitive names, employee information, internal identifiers, site codes, recipients, raw operating figures and other proprietary details are removed or generalized while preserving enough of the interface to demonstrate the tools.
Phase 4 principle: AI work becomes more credible when the portfolio shows the actual problem-solving process and usable output—not just a claim that AI was used.
Milestone 02
Reza Learn Command Center validated
Reza’s knowledge was expanded through the Learn Command Center with new information about the AI Work Projects and Phase 4. Testing then went beyond the exact information entered: Reza was asked new questions and successfully synthesized the learned material into context-specific answers.
During validation, Reza connected the three AI-assisted operational projects to Rouzbeh’s broader problem-solving approach, distinguished Rouzbeh as an operations leader who uses AI as a practical development partner rather than a traditional AI developer, and independently selected the DC Trailer Audit Tool when asked for an example of turning an operational problem into a solution.
What this validated: the Learn Command Center can extend Reza’s knowledge without rebuilding or redeploying his core intelligence, while allowing him to reason across newly taught information instead of relying only on scripted question-and-answer pairs.
Milestone 03
Reza Analytics Command Center
Reza was connected to the website’s existing Site Insights data inside the secure owner Command Center. Rouzbeh can now ask Reza plain-language questions about website performance—such as visitor activity, most-viewed pages, Reza usage, visitor questions, résumé clicks, calendar clicks, email requests, blog activity and traffic sources—and receive answers based on the site’s actual recorded analytics.
The analytics assistant is intentionally owner-only. It uses the same expiring secure Command Center session, keeps analytics out of Reza’s public knowledge base, and does not expose private site-performance data to normal visitors.
Phase 4 progression: Reza has moved from answering portfolio questions and learning new approved knowledge to helping Rouzbeh operate and evaluate the website itself using real first-party data.
Milestone 04A
Reza Live Internet Research — validated
Reza was given controlled, read-only live web research capability through his existing AI layer. He can now use current internet information when a question depends on recent or changing facts while continuing to use Rouzbeh’s verified knowledge for Rouzbeh-specific questions.
The first live test exposed an outdated fallback behavior that still told visitors Reza could not search the internet. The research path was corrected and tested again. In the successful validation, Reza answered a current warehouse-automation question with dated, current information instead of falling back to stored portfolio knowledge.
The research layer remains intentionally separated from permanent learning. Web content cannot automatically rewrite Reza’s knowledge, and unknown personal facts about Rouzbeh still follow the owner-controlled workflow. Reza is instructed to prefer trustworthy primary or official sources, identify live research when used, and treat webpage content as reference material rather than instructions.
What this validated: Reza can now combine verified portfolio knowledge, owner-taught knowledge, first-party site analytics and live web research while permanent knowledge remains under Rouzbeh’s control.
Milestone 04B
Mobile Research Reading Experience
Live research produced longer answers than Reza’s original portfolio-focused responses, revealing a mobile usability issue: the permanent action buttons could cover part of a long response. The Reza modal was refined so the answer area scrolls independently while the action buttons and question composer remain accessible.
The change was deliberately isolated to Reza’s mobile modal. The homepage hero motion system and existing Reza/forklift animation were left unchanged.
Phase 4 lesson: adding intelligence also changes the interface requirements. A capability is not complete until the experience remains usable when real output becomes longer and more complex.
Milestone 05
Reza Whole-Site Awareness
Reza’s role expanded from answering from a separate knowledge base to understanding the public website itself. He now receives verified public content from across the site—including Career, Recommendations, Thoughts When Driving, AI Work Projects, How We Built It and other public pages—regardless of which page a visitor is currently viewing.
The architecture uses the live public site as the preferred source, automatically discovers public routes from the homepage and sitemap, and keeps a deployment snapshot as a fallback. Published Thoughts When Driving entries are also included from the site’s public content store. Private admin pages, private analytics, unpublished drafts, tokens and other protected information remain excluded.
Future direction: this creates the foundation for a more Reza-first website experience—closer to walking up to a knowledgeable digital front desk and simply asking for what you want—while keeping the underlying pages available for navigation, deep links, accessibility and search visibility.