
OfferPilotV2
Desktop-first, local-first job-search command center. Multi-agent pipeline with Scout / Tailor / Researcher subagents and human-in-the-loop approval.
Years in the L2 queue at AI-forward SaaS — sustained exposure to where customers hit walls. I build the local-first tools support, CS, and SE teams actually need, eval-driven and shipped.


Desktop-first, local-first job-search command center. Multi-agent pipeline with Scout / Tailor / Researcher subagents and human-in-the-loop approval.

Local-first AI triage for B2B SaaS support — classify, retrieve KB, draft citation-grounded replies, suggest macros. Eval-driven.

Single-page Streamlit dashboard a CSM opens every Monday. Documented v1→v5 prompt iteration with held-out scenarios.

MLB pick recommendation system — gradient-boosted models plus grounded LLM rationale plus two parallel Claude agent pipelines.
Customer-facing rigor combined with the engineering instinct to ship AI features that actually work — not demos, not slideware.
Customer trenches
Sustained exposure to where customers hit walls at AI-forward SaaS. I know what breaks because I've worked the queue, not because I read the support docs.
AI tooling
Local-first agentic pipelines, multi-LLM provider abstractions, eval-driven prompt iteration. Tools I've shipped: triage, sourcing, account pulse, job-search orchestration.
Solutions engineering
Comfortable across the stack — Rust, TypeScript, Python, SQL. Security+ certified. FINRA 7/63 (lapsed) for financial-services context.