I turn AI prototypes into products people actually rely on in production
Senior Technical Program Manager repositioning toward AI Product Management — leading initiatives end to end, from roadmap and stakeholder alignment to the retrieval and agent systems that ship real outcomes, not demos.
Drawn by J. Cohen · Portfolio 2026 · Scale NTS
Agent and retrieval systems, shipped.
Prompt Dojo
→A gamified, belt-ranked prompt-engineering trainer, built as a single dependency-free HTML file
AI News Digest
→A scheduled research agent that reads, verifies, and remembers — so the daily AI news summary never repeats itself or pads a quiet day
Job Hunter
→A resume-driven job search skill that deep-matches listings on a weighted rubric and discloses coverage gaps instead of hiding them
Graph-Engineer: DAGs of Sub-Agents
→A Claude Code skill that turns a prompt into a validated graph of parallel sub-agents instead of one long serial run
Jira Triage Agent
→An LLM agent that reads tickets, searches a shared knowledge base, and cross-references logs before an engineer ever opens the ticket
From technical program management into hands-on AI product work.
I didn't wait for a title change to start doing AI product work — I started building. The projects above weren't assigned to me; they're what nights and weekends looked like while I found out, hands-on, whether I could actually ship this stuff or just talk about it.
That same instinct carries into my day job as a Senior Technical Program Manager, where I've built a shared RAG knowledge base on pgvector, wired agents into real systems of record through MCP, and shipped an autonomous investigation agent I call Jira-Bug-Investigation. Different stakes, same approach.