Senior Workplace Coordinator at Airtable, managing the 300-person San Francisco headquarters and supporting four global offices. I taught myself AI-agent development and turned my own team's busiest workflows into production automation: visitor management, badging, ticketing, and events. Along the way I won Airtable's company-wide builders prize.
Every project below started as a manual workflow I personally ran, then rebuilt as software. Status labels are honest: Live means running in production today; Pre-launch and Prototype mean exactly that.
Multi-office visitor management was hours of daily copy-paste between the visitor platform, Slack, and building-security portals. I built a fleet of agents that ingests visitor data via the Envoy API, posts daily digests to Slack, and completes building-portal registrations end-to-end through authenticated browser automation, with verification at every step so nothing is submitted unchecked. Running in production daily since April 2026.
Workplace tickets were answered by hand, one at a time. I first deployed and trained a vendor AI help desk, learned its limits, then built our own in-house agent grounded in the company's Atlassian knowledge base, so answers come from our real documentation instead of guesses.
Requests arrived scattered across Slack DMs, hallway asks, and email, untracked and unreportable. I built the team's ticketing system in Airtable with Slack and email intake, turning every request into a tracked, measurable queue. This is the system the AI agent now deflects tickets from.
Badge-access events lived in a vendor silo with no way to answer "who actually uses the office?" I architected a 17-table platform ingesting access-control webhooks with deduplication and idempotency safe under concurrent delivery, reconciled against the employee directory. A monitoring agent watches the pipeline and caught a silently dropped webhook in production.
The legacy badge-tracking base had drifted from reality: I measured 85% record drift and badges stranded with departed employees. I replaced 15 legacy automations with a self-correcting request/loan system and native approval interfaces.
Company events ran on tribal knowledge. I built the reusable machinery: a city-parameterized venue-outreach pipeline with reply tracking and no-double-email guards, an RSVP/check-in/feedback system with QR distribution and AI-summarized post-event ratings, and attendance analytics correlating 484 prior-year RSVPs against comms touchpoints.
Vendor compliance lived in inboxes. I built an 8-table system tracking 22 vendors and 64 contacts across four sites: certificate-of-insurance expiry against per-building requirements, service agreements per office, and budget-variance formulas tied to procurement.
Work-anniversary recognition meant manual gift orders every week. I built an agent that handles it end-to-end, designed safety-first: stage-only purchasing, read-only scoped tokens, and PII-safe logging. Validated end-to-end on synthetic data; in pre-launch.
Hosting onsite interview candidates took coordinated manual work across recruiting, calendars, and the visitor system. My working prototype chains recruiter intake into auto-created calendar invites with room booking, visitor invites with NDA delivery, and auditable minimal-PII confirmations. Phased rollout planned.
Simultaneous summer picnics in four cities, ~150 guests per site. I built the RSVP system with dietary rollups and a live dashboard, contracted venues, and ran a 26-post communications kit against a structured posting calendar.
The systems above aren't side projects. They automate work I own, and that operator context is why they get adopted.
Manage in-office operations for Airtable's San Francisco headquarters; support NYC, Austin, and London remotely: building relationships, badge/access, and visitor programs.
Own the company's flagship internal events (picnics, holiday parties, hack/build days, and 800-person all-hands where I draft the exec decks, run AV, and have MC'd) on a ~$300K annual budget.
Led two company-wide food-program transitions (pantry and daily lunch, each a $1M+ annual contract) from vendor selection through multi-office rollout. Manage 8+ vendors day to day: performance, scopes of work, on/offboarding.
Ran Airtable's Austin office end-to-end (65–100 people, 20+ hires onboarded monthly), then fully transitioned my successor and helped train the NYC coordinator remotely.
Founding Culture Ambassador, advanced to the company Culture Committee alongside the Director of Employee Experience and Chief People Officer: company-wide culture, the ambassador program, and six ERGs.
Taught myself Claude Code, OpenAI Codex, and AI-agent development, then applied them to the workflows I knew best. Operator first, builder because of it.
Four rules that decide what I ship and what I refuse to ship.
Every system started as my own manual workflow. I don't guess at requirements. I've done the job the software replaces.
Least-privilege tokens, stage-only spending, PII-safe logging, human approval where it matters. Agents earn autonomy; they don't start with it.
Idempotent ingestion, verification before submission, and monitoring agents watching the pipelines; one caught a silently dropped webhook in production.
Live means live. Pre-launch means pre-launch. The same discipline applies to metrics: estimates are labeled as estimates.