Build · Ship · Adopt

I build AI products end-to-end — and get them adopted where it's hardest.

I'm Ari Rosenberg. I shipped a full-stack AI product solo, concept to live. By day I build LLM systems inside federal health.

View projects → Resume (PDF) Get in touch

About

I design and ship AI products end-to-end — the problem, the product, the prompts, the code, the infrastructure. Most recently ACT Journal, a full-stack AI journaling app I took from concept to live, solo.

By day, 15+ years at Booz Allen Hamilton across federal health. I build LLM systems where shipping means earning trust in governed, restricted environments. Right now: an LLM-powered content pipeline assessing thousands of pages for a major federal health agency.

How I build: evals before features, cost as a first-class design constraint, and the prompt layer treated as product — versioned and tested like code.

Selected Projects

Featured · AI Product · Solo build

ACT Journal

An AI journaling app grounded in Acceptance & Commitment Therapy. You define your values, write daily entries, and the AI reflects your writing back against what matters — surfacing patterns, tracking behavioral shifts, and closing the loop between insight and action.

Built the full product solo — concept, UX, frontend, database, and the entire AI layer.
Multi-provider AI proxy (Claude, GPT, Gemini) with a persistent "working formulation" that keeps the AI calibrated to you across sessions.
Cut per-session AI cost ~97% via prompt caching, three-tier model routing, and context tiering.

React 19SupabasePostgres + RLSEdge FunctionsClaude / GPT / GeminiPrompt caching

Live site ↗ Technical deep dive → Interactive tour ↗

Current · Enterprise AI · Federal Health

Content Intelligence Pipeline

An LLM-powered content-governance system for a major federal health agency — turning a sprawling legacy SharePoint estate into a scored, prioritized modernization backlog and the foundation for a shared-services portal.

Architecting the full pipeline: scripted content inventory and extraction, deterministic quality scoring, and local-LLM assessment across thousands of pages.
Designed for a restricted government environment — local models (Ollama), with content never leaving the boundary.
Roadmap: agent-based content monitoring informed by real user queries, with assessment that improves over time.

Enterprise / government work — details shared on request.

Content Intelligence

Background

15+ years at Booz Allen Hamilton across federal health — currently AI & platform development for a major health agency, and before that analytics and AI product development on enterprise-scale health programs. PMP-certified.

Full résumé (PDF) →

Education

B.S., Finance & Information Systems

University of Maryland, College Park

Minor in Rhetoric · Full-Tuition Scholarship

Let's talk.

Open to roles, collaborations, and conversations about building with AI. The fastest way to reach me is email.

rosenberg.ari@gmail.com LinkedIn GitHub