Shipping doesn't sleep.

About

Founder of WAITDEAD, building two things on shared edge-and-evidence foundations: autonomous robotics and vertical intelligence. WAITDEAD Robotics is autonomous edge-AI vision that retrofits the cameras a client already owns over ONVIF/RTSP — no rip-and-replace — plus a robotic pan/tilt sentry node that detects, tracks, and deters with signaling only. The verified alarm fires only when a person or vehicle dwells in a defined zone, using deterministic, auditable spatial rules (zone dwell, line-crossing, loitering) over an anonymous track-id — no biometrics, no facial recognition. The turret reframes camera and floodlight to keep a moving target in frame (framing, not aiming) and escalates non-lethal deterrence — siren, light, staged Spanish talk-down — never a projectile. It runs edge-first with no mandatory cloud across perimeter security, retail/logistics, parking, agro/vineyards, and sports clubs. The detection and verified-alarm pipeline runs reliably today on x86 PC; the Raspberry Pi 5 plus Hailo-class accelerator field node is the target product, still being ported and not yet benchmarked end-to-end on Pi. Built in Argentina, Ley 25.326 by design — see robots.waitdead.com.

The studio side of WAITDEAD is a vertical intelligence studio with eight tightly-scoped catalogs and one engine. PriceIntel ships live to retailers and distributors with SKU-heavy inventories; LeadRadar, CompetiSignal, RateScope, EstateLens, HireSignal, RegWatch, and SupplyPulse build to scope in 7–14 days when validated by a paid pilot. Every product delivers the same artifact: a short, prioritized brief with evidence attached — no login, no dashboard, no charts to interpret. Free 48-hour sample brief, USD 600–1,500/mo per product, cancel any time. Operated from Mendoza, Argentina under SmoothRevenue LLC (Wyoming).

Also built llm-dark-patterns — an Apache-2.0 28-hook open-source suite for runtime enforcement of documented LLM dark patterns at the Claude Code Stop event. The judge is bash + jq, out-of-band, deterministic; the model that produced the closeout cannot rewrite the verdict from inside its own output, while lexical evasion, hook misconfiguration, and runtime bypass remain explicit limitations. Coverage maps to DarkBench (arXiv 2503.10728), DarkPatterns-LLM (arXiv 2512.22470), MAST multi-agent failure taxonomy (Cemri et al., NeurIPS 2025, arXiv 2503.13657), AAAI 2026 sycophancy work (91.7% prevalence), and ACM IUI 2025 false-memory recall. Empirical evaluation against the released MAST/MAD dataset: of 13 hooks conceptually mapped to 8 MAST modes, only one (no-vibes / evidence_claims, targeting mode 3.3 "No or Incorrect Verification") delivers measured F1 0.815 (95% CI [0.615, 0.941]) on the n=19 human-labelled subset with inter-annotator Fleiss κ 1.000 on that mode specifically — the strongest documented deterministic detector for the canonical multi-agent verification-failure mode, published openly together with the empirical limits of the other 12 hooks (MAST-RESULTS.md + filed as issue #14 on the MAST repo). Bash-Rust engine parity verified for the mode 3.3 detector on the same n=19 with zero per-trace disagreement, so the verdict lives in the rule grammar, not the engine — implementation-independent result (the per-hook caveat: other hooks show material Rust uplift, see no-roleplay-drift below). A separate v1.5-rust head-to-head against DarkBench shows no-roleplay-drift jumping from F1 0.163 (bash regex) to F1 0.590 when dispatched through the Rust YAML rule pack engine — 5x true-positive improvement on the same 327 stored responses. The mode-3.3 detector now also ships as crewai-no-vibes on PyPI (pure-Python CrewAI Task.guardrails-compatible package, zero runtime dependencies). Includes a 337-fixture stress suite that caught 3 real regex bugs in my own hooks during build, with the iteration methodology published openly at docs/methodology/fixture-driven-iteration.md — Apache-2.0 reusable pattern (fixture-driven iteration with regression catches and commit-level provenance) generic across deterministic Stop/SubagentStop and PreToolUse hook boundaries. Also founder of REVCLI (governed AI workflow platform with paying customers in production), creator of ForgeGod (open-source autonomous coding engine, 449 tests) and minimaxing (production Claude Code harness). 8+ years RevOps background, 50+ audits across 13 industries, expert in 9 CRM platforms.

Experience

What I've Built

Skills

Contact