Aviation Platform & ML — 2024–26
Aircraft Executives.
An aircraft-sales operating system built on JetNet data — market intelligence, a print-grade report engine, a client portal, and a machine-learning program with the honesty machinery to grade itself.
- Role
- Head of Product
- Folio
Head of Product across the full surface: brand identity, market collateral, and the platform at app.aircraftexecutives.com. What began as spec sheets and model reports grew into a CRM, a client portal, a market-intelligence desk, and a fleet of about a hundred scheduled jobs — roughly nineteen thousand commits in fifteen months from a very small team working with agents.
Reports were the product. A PDF engine rendered spec sheets, model reports, monthly owner updates, and quarterly market reports with an exact on-screen preview of the printed page. Market Pulse tracked every model the firm cared about — fleet size, inventory, average ask, days on market, data freshness — fed by JetNet transaction history, Flightradar24 utilization, and a hand-curated research sheet that no competitor could buy.
Under the reports, a four-model machine-learning portfolio: per-tail market likelihood, deal closure, reply prediction, and outbound effectiveness, trained on Modal and evaluated with walk-forward backtests, leakage checks, weekly champion-versus-challenger runs, and drift gates. Scores surfaced on a lead map, in prospect lists, and in daily revenue briefings.
The workflow layer shipped with its control surface: a client portal with embedded DocuSign signing and self-serve onboarding, inspection and airmen registries, an outbound kill switch, and a gated-writes proposal queue so that every AI action was reviewed before it touched a deal.
The same machinery delivered the verdict. In August 2026 an honest backtest put the champion model at 0.70 AUC and 1.46× lift, not the 0.96 and 9× the early metrics claimed, and a signal-ceiling study found no early-warning signal for sellers in any data available at the budget. The sponsor withdrew that week. Six packages left as a private toolkit — PDF engine, unified cache, API resilience, MLOps kit, job primitives, design tokens — along with the rule now written at the top of every new venture: probe the moat before building the castle.
