Automation, backend, and a bit of hardware.
The flagship build has its own case study. Everything below is real, running (or ran) code — described with the numbers I actually have, not the numbers that would sound better.
Hardware-accelerated Raspberry Pi camera stream
A Pi Zero 2W livestream server was software-encoding MJPEG in Python — pegging the CPU and cooking the board. Rewrote it to shell out to rpicam-vid and let the VideoCore GPU do H.264 encoding instead, served over a small stdlib HTTP server as a systemd user service. Same camera, same stream, radically different resource profile.
AI lead-qualification & drafting engine
An n8n workflow that takes an inbound lead, scores it against a fit rubric using the Gemini API, and drafts a first-pass response — logged to Airtable with the full score breakdown visible, not a black-box number. Nothing sends without a human clicking approve; the point was speed on the good matches and an automatic no on the bad ones, not a fully autonomous pipeline.
Built to replace a manual triage process that didn't scale: every incoming opportunity used to get the same amount of attention, whether or not it was worth pursuing. The engine reads the same signals a human would, just faster and every time.
Live Kanban dashboard over Airtable's REST API
Airtable's own Kanban board is an Interfaces feature gated behind a paid tier. Rather than pay for a view, I built the view: a small Flask app that reads and writes the underlying table directly through Airtable's REST API — drag-and-drop-free, but instant, and it keeps Airtable's own grid view in sync since it's hitting the same data.
A small, honest example of the instinct I bring to client work: when a SaaS tool's own feature is behind a paywall or missing entirely, the fix is often a 200-line script, not a subscription upgrade.
Voice-agent intake for host onboarding
Property hosts described their listing over a phone call instead of filling out a form. A Bland AI voice agent ran the conversation, and the responses were parsed and written directly into the platform's structured fields — turning an onboarding step that used to require a form (and usually a follow-up email to fix it) into a five-minute call.
Paired with a second pipeline: once a proposal was accepted, a Google Doc contract auto-generated and an Eleven Labs voice narration of the house manual was attached — the same "let the system do the talking" idea applied to onboarding's other end.
Multi-channel message proxy router
A lightweight Flask gateway on PythonAnywhere sat between the platform and its notification channels, intercepting webhooks and routing each message to SMS (Twilio), email (SendGrid), or in-app delivery based on that user's actual preference — replacing what had been a Zapier subscription with a service that cost a fraction as much to run.
Where this came from.
Split Lease — Lead Full-Stack Developer / Technical EA
PropTech startup, recurring NYC rentals · 2 years · 3,356 verified hours across 104 contract weeks
Worked directly alongside the founder, leading the technical side of the business: the Bubble-to-code migration on the case study above, plus everything on this page. Went from no prior coding background to running the engineering side of a live startup within one long-term engagement.
Thankful AI (acquired by Gladly) — AI Training Assistant
Enterprise AI customer service platform · 4 years
Classified and labeled customer support conversations at volume to train the platform's models — the ground-floor view of what makes an AI system actually useful in production, before the automation engineering came along.
Something similar on your plate?
Migration, automation, or an AI workflow that needs building — happy to talk through it.