Applied ML · Digital Experience · Tesla
Agentic Workflows · Traditional ML · Software Engineering.
Click a node to light its neighborhood — stack and paradigms follow.
Production systems I led or owned. On-call across dispatch, booking, capacity, slots — NA & EMEA.
Geo demand + predictive models into routing and booking.
ML · clustering · demand · Airflow
Capacity-buffer trials into Voyager — treatment vs control, field-validated.
Experiment · A/B · demand-aware routing
Dispatch engine: prioritization, Hex7, ALNS-style routing search, prefetcher scale, on-call harden.
Systems · ALNS · PrioritySchedulingV2 · Hex7 · on-call
Scheduling assistant for Service & Mobile ops — policy, skillset, RAG tools.
Applied AI · tools · RAG · NLU
Cancellation-aware overbooking + Airflow rewrite across 400+ locations.
Policy · cancel risk · WIP · Airflow
Supporting systems and judgment calls behind the signature outcomes — lower weight, still shipped.
Skills + ReAct agents and infra so on-call bots turn into productive engineering hours — not just alert spam.
Applied AI · ReAct · skills · agent infra
Multi-tenant agents from Jira — planning, codegen, review for 6 IT projects; Tesla One Mobile debug.
Platform · multi-tenant agents · Jira
Modeled Preferred Slots ROI (~4% ceiling) and staffing spikes; killed low-upside work so Buffer Buster–class levers kept eng time.
Judgment · go/no-go bars · roadmap influence
Open-slot-by-skillset pipeline — demand/supply base for work-mix planning across field capacity.
Analytics · open slots by skillset · work-mix
Experience paradigms: Velocity, 80/20, 0 → 1, 1 → 100.
Ship, measure, harden — short loops over perfect plans.
Kill low-ceiling work; put force on the lever that moves the metric.
Jarvis, Robodev, agent skills/infra — zero to working product.
Voyager scale, booking across 400+ sites, agent adoption under load.
Preferred Slots ~4% impact cap → deprioritized. Staffing spike killed before months of eng burned.
Voyager: 500s 24→4, RMQ 300KB→3KB, stale-connection recovery — permanent fixes, not heroics.
Buffer Buster design→trial→field. Jarvis backend→7k tool calls. Booking pipeline 6h→30m.
Hard ROI bars. Kill weak paths. Push roadmap toward capacity levers that compound — not sole org owner of the roadmap.
Onboard eng + interns onto dispatch, booking-rate, and capacity systems so they ship with production standards — pairing, review bar, shared on-call load.
Tier-3 primary on dispatch (Voyager), booking-rate, capacity analytics, and slots pipelines — feature/project ownership with clear bounds, not sole TL of Digital Experience.
Short form. Signature outcomes above. Here: scope growth with ownership bounds.
Expanded ownership on production agents and internal platforms — Jarvis backend (7k+ tool calls, 700+ ops users) and Robodev adoption (Jira multi-tenant path for 6 IT projects; Tesla One Mobile observability). Multi-system ownership with clear consumers; not org-wide platform TL.
Core decision-systems loop: demand/throughput ML (~+20% NA Mobile; pilot backlog ~−40%), Buffer Buster capacity trials (project lead; est. $12.3M trial window), Voyager reliability + scheduling features (feature owner + tier-3, not sole Voyager TL), booking-rate pipeline (−91% runtime) and cancellation-aware overbooking. Depth in Work above.
Hypothesis-driven demand experiments (sparse/dense regions) cutting average drive times; faster vehicle-routing APIs via dynamic regional localization; HPC simulation of scheduling systems; visualization of zones, routes, throughputs, and drive times.
Graduate algorithms, supervised & unsupervised ML, NLP, data management, DBMS, cloud (AWS).
ML clustering on gene sequences; ~40% cumulative diagnostic coverage; outlier pipelines from 510k filings; end-to-end primer-design pipeline for fast-evolving analytes.
Offline content download backends with Protocol Buffers (−30% overhead); offline viewing UX on Oppia Android.
Dynamic partitioning on Android Q update engine; In-Device-Diagnostic app for boot/update crashes — field-facing systems software for Sony Mobile.
Advanced eng math, discrete math & probability, linear algebra — signal, noise, systems under constraint.
Decision systems, logistics/ops ML, production agents, multi-surface ownership with hard boundaries. Fremont / Bay Area. Not foundation-model research — building toward Staff, not claiming the title.
Off the laptop: Pi automations, long runs, trails, landscape photos.
Capacity, routing, booking policy, ops agents, production reliability — blast radius required. Senior+ / Staff-track scope; clear ownership bounds.
IoT controllers, Pi setups, workspace automation — same build muscle, different medium.
Distance running, hiking, landscape photography. Best debugging often starts offline.