Applied AI systems
I ship applied AI that operators and enterprises can run — harnesses, agents, and personalization systems under real constraints, not slideware.
Forward-deployed delivery on the floor. Enterprise AI programs at portfolio altitude. Same mind: constraints, human gates, something runnable on Monday.
Proof pillars
Architecture and outcomes only. Client names anonymized. No screenshots of private systems.
01
Built a UI-agnostic agent harness that extracts structured work-order fields from multi-page industrial intake documents via vision, then writes to a tenant ERP contract — only after a technician confirms.
Separated the harness brain from a mobile-first intake UI, with a session state machine, knowledge-base lookup, and stub/live adapters so the loop runs offline-first. The product is the extraction → confirm → write path, not chat.
02
Built alternate-part validation tooling over a large OEM technical-manual corpus with three-way cited verdicts — confirmed, contradicted, or undetermined — and auditable per-claim method trails.
Authored a normative interchangeability knowledge base (OEM tokens, regulatory codes, legend precedence) and treated undetermined as a first-class outcome. Controlled manuals stayed offline; the system carries code and docs only. Semantics beat more LLM calls.
03
Built a print-faithful multi-page document viewer with deterministic PDF export and an embedded document-aware sales agent — for a commercial interiors / building-services manufacturer.
Print fidelity was a product requirement, not a nice-to-have: page-hash cadence against source art, lead capture on the edge, and messaging constraints treated as requirements. Rebuilt mocks only — no client screenshots here.
At scale
Large U.S. network operator · personalization & GenAI org · 2019–2026. Themes and altitude only. Exact dollars withheld pending confirmation.
Built production ML ops and governance for a large personalization portfolio — intake, monitoring, holdouts, Finance-validated valuation — with a lean operating team.
Directed GenAI productization for marketing ops: in-house copy generation and the org’s first automated legal/compliance scoring path for scaled campaign content.
Scaled outreach and intent personalization programs with holdout-measured impact; partnered marketing, digital CX, and technology for mainline adoption.
Deployed retention and dynamic-offer models across mobile and home connectivity segments; treated measurement design and guarded outcomes as first-class.
Earlier: built real-time marketing and CRO experimentation capability that later funded AI credibility — from pod execution to Sr. Mgr / Assoc. Director altitude.
Prototyped guided, conversational account-review experiences for executive buy-in — intent taxonomy, plan optimization, live bill-impact — concept demos with mocked data.
Range
How I work
The workflow is the brief — hangar floor or marketing ops. Monday has to work without me in the room.
Open kits
How I ship decisions — HTML-first DX and design-to-LLM tooling. Same rules for humans and codegen.
About
I’m Reed Garrett — I ship applied AI under production constraints. Sessioned harnesses and document agents on the floor; personalization, GenAI ops, and model governance at enterprise scale. I unify cross-team delivery and leave systems people can run without a demo babysitter.
Work shown here is rewritten for public use: industry framing, architecture diagrams, qualitative outcomes. Private catalogs, employer brands, and client repos stay private.
Public boundary: anonymized clients by default · no unverified metrics · no private repo links. See PUBLIC_BOUNDARY.md.