Oren Segal

AI engineer and founder of Shelfie, an iOS app that builds a kitchen inventory from a camera walk-through of the fridge and pantry, then plans meals, drafts grocery lists and tracks what is about to expire.

Claude

Code agents wrote most of the code across 2,100+ merged PRs, so much of the work is reviewing that code and building the tests and CI checks that catch where it falls short. Before that, five years as a Senior Data Analyst at Reshet 13, an Israeli TV network, building Python and SQL pipelines and forecasting models.

Founder & Lead Engineer · vision-based AI kitchen app for iOS, in TestFlight beta - Shelfie - New York

(2025)

• Designed the core interaction, a continuous camera tour of the fridge, pantry and freezer that recognizes food as the user pans. Confident items save automatically and uncertain ones wait for a quick correction, so stocking the kitchen takes one walk-through instead of manual entry.
• Connected the rest of the cycle to that inventory: receipt scanning, expiry tracking, meal planning, leftovers, and a grocery list drafted from the week's plan for the user to confirm.
• Own the architecture, infrastructure and budget: a modular Swift/SwiftUI codebase on a Supabase backend (Postgres, auth, storage, edge functions).
• Built the vision pipeline for real kitchens, where frames are shaky, the light is poor and labels are hard to read. Blurry frames are dropped and dim scenes prompt for more light. On-device recognition (Apple Vision, OCR, barcodes) handles what it can, and only what it can't resolve goes to a cloud model, which keeps AI cost low.
• Put guardrails on every model call: one gateway with spending limits, caching and a circuit breaker, plus a CI check that stops the client from calling providers directly. When an agent's cost estimate for scanning proved about 34x too low, replaced the estimates in the code with measured billing.
• Wrote the recommendation logic for recipes and weekly meal plans as explicit scoring on expiry, what is on hand, diet and taste, so each suggestion comes with a reason the user can read.
• Set up the Claude Code harness the agents work in and kept a human in the loop on merges: the merge script stops for me whenever a review asks for one. CI targets the ways an agent can pass checks without doing the work, such as a test run that executes zero tests. Those checks are tested too.

Senior Data Analyst - Reshet 13 - Tel Aviv

(2018 - 2023)

• Found underpriced primetime ad slots with a regression model, then worked alongside the commercial team until the new prices went live. Revenue rose 17% in FY2022.
• Developed the network's main viewership forecasting model, which became the standard input for leadership's content acquisition and greenlight decisions.
• Automated more than 70 weekly Excel reports as Python and SQL pipelines, saving about 70 hours of manual work a week.
• Mentored four junior analysts through the team's full move from Excel to automated pipelines.

Chief Growth Officer & Board Member (volunteer) - Cohort NYC (cohort.at) - New York

(2026)

A NYC learning community built on weekly coworking and regular events, with 1400+ Luma subscribers and 1100+ unique participants so far.

• Built a pipeline that loads Luma event registrations into Postgres and ties each one to the link or referral that brought it in, showing which channels worked. Free subscriptions and event attendance doubled over 75 days.

Bachelor of Science - Information Systems - The Academic College of Tel Aviv-Yaffo (2014 - 2018)