Ezra Apple

B.S. in Electrical Engineering and Computer Science

• Helped architect and launch Replo’s new AI platform for e-commerce through closed beta, open beta, and general availability, with work spanning reusable templates, agent behavior, and live page previews.
• Redesigned Replo Agent’s persistent memory around compact user and project context, on-demand retrieval, and automatic curation, and shipped a user-facing memory browser.
• Designed and ran agent evaluations of context and instruction changes using customer-derived and synthetic editing tasks, measuring task correctness, cost, and latency.
• Led consolidation of Replo’s internal agent skills and development workflows, making engineering, design, and support knowledge discoverable and reusable across coding agents.

• Co-founded AI video translation startup, accepted into UC Berkeley SkyDeck Pad-13 incubator program.
• Owned end-to-end MVP for AI-driven video translation across 25+ languages, spanning authentication, media upload, processing orchestration, and launch site.
• Ran closed beta with 10+ content creators, translating feedback into product refinements and go-to-market validation.
• Architected a serverless media-processing pipeline on AWS (S3, Lambda, Step Functions, SES) with fault-tolerant retries for long-running translation jobs.

• Applied random forests and Bayesian optimization to ion-source diagnostic data, modeling accelerator performance and exploring configurations to maximize ion output.
• Presented findings to senior researchers supporting heavy-element synthesis research, including the element 120 program.

ShoutOut | Local-First macOS Dictation App

• Built and shipped a local-first macOS dictation app in Swift that inserts transcribed speech into other apps, with signed, notarized releases and automatic updates.
• Designed a low-friction interaction model with hold-to-record, double-tap hands-free mode, menu bar controls, recording overlay feedback, and first-run permission onboarding.
• Implemented swappable on-device transcription engines using Apple Speech, Apple Dictation, and WhisperKit/Core ML, with dictionary-aware cleanup and spoken-punctuation formatting.