Hey 👋 I’m Sena Oz.

Software Engineer

I am a software engineer who cares whether systems actually meet people's needs. I hold an M.S. in Software Engineering and a B.S. in Management Information Systems from Boğaziçi University, and I work at Maestra.ai on AI speech products. I am interested in how developers and researchers retrieve, verify and come to trust AI-generated code and information.

LinkedInEmailGitHub

Experience

  1. Jul 2024 – Present

    Maestra.ai

    Software Engineer

    • Full-stack engineer on AI transcription, dubbing and real-time captioning products (React, Vue, Node.js, Python; Gemini, Soniox, ElevenLabs).
    • Built an LLM translation pipeline with provider fallbacks and recovery from malformed model output, and moved voiceover generation onto Google Cloud Tasks with retries and rate limiting.
    • Built user controls for steering live AI captioning and translation: custom vocabulary and word replacements for speech recognition, and per-language glossaries injected into Gemini translation prompts with fuzzy whole-word matching.
  2. Aug – Oct 2023

    California State University

    Software Engineer Intern

    • Built an LLM-powered (OpenAI API) platform that gives students feedback on their code and collects that feedback data for model improvement (React, TypeScript, Tailwind CSS, Node.js).
    • Worked with faculty from design to deployment.
  3. Nov 2022 – Nov 2023

    Orion Innovation

    Junior Software Engineer

    • Worked on a client project for a U.S. genetic-testing company within a 1,200+ developer organization.
    • Developed reusable, accessible (WCAG) React/TypeScript components for its web and mobile portals, documented in Storybook.
    • Added visual-regression and unit tests to keep the UI library stable.
  4. Mar 2021 – Jan 2023

    YetGen

    Web Application Dev., Team Lead

    • Led development of a learning management system for 3,500+ learners that replaced a third-party platform, built from scratch over two years as the team grew past ten people.
    • Designed and implemented its authentication module (JavaScript, Next.js) and took part in designing its database.
    • Built it around WordPress, Airtable and Moodle with custom Next.js parts so the non-profit could run it without outsourcing, and owned its AWS and Azure infrastructure.
  5. Jul – Aug 2022

    Univerlist

    Front-end Engineer Intern

Projects

  1. The Hive — Community Time-Bank Platform

    • Full-stack web platform enabling users to exchange services using time credits.
    • Built with React (TypeScript), FastAPI, and MongoDB.
    • Features include service listings with geospatial search, a community forum with WikiData-powered semantic tagging, real-time chat, 'time-credit transactions', and an admin moderation panel.
    • Deployed via Docker and GitHub Actions CI/CD to a cloud host.
    • Companion Android app built with Kotlin and Jetpack Compose.
    GitHub ↗

  2. Scholar Radar: Academic Paper Recommendation and an LLM Reranking Audit

    • Built Scholar Radar as a Graduate Research Project at Boğaziçi University: a recommender that monitors followed institutions on OpenAlex for new papers and lists the Boğaziçi researchers who work on each paper's topic (7,822 cleaned works, 3,648 researchers).
    • The research question was whether reranking embedding-retrieved candidates with a small local LLM improves recommendation.
    • A first benchmark suggested it did (MRR 0.84 vs. 0.51); an audit of that evaluation found nine problems, chief among them that cited papers were always shown to the LLM first.
    • On a rebuilt, leak-free benchmark (118 test queries, 13 methods, five shuffle seeds, Holm-corrected tests), Qwen3-Embedding-8B reached 87.0% R-precision against 43.9% for listwise Llama 3.2 3B, which largely copied its input order: the same model scores MRR 0.95 with cited papers first and 0.33 with them last.
    • The conclusions held on an 11,967-paper corpus with unit-tested leakage rules and on expertise ratings from 58 researchers, so the LLM stage was removed and the system now uses a single embedding stage.
    • Paper draft: "Do Small Local LLM Rerankers Help Scientific Paper Recommendation? An Audit and Re-evaluation with Citation and Human Expertise Labels."
    • Stack: OpenAlex, FastAPI, PostgreSQL/pgvector, llama.cpp, React, TypeScript, Docker.
    GitHub ↗

  3. Code Generation and Quality Checking using LLMs

    • Term paper for SWE 577 at Boğaziçi University, advised by Assoc. Prof. Atay Özgövde.
    • Analyzed how code-generation benchmarks (HumanEval, MBPP, SWE-bench, SWE-Lancer) serve as proxies for developer trust, arguing that test pass rates miss semantic correctness, maintainability and usability.
    Live ↗

  4. Ontario International High School: Website and Admissions Portal

    • Sole engineer for the school's bilingual (TR/EN) website and parent, student and teacher portal (React Router SSR, Node.js/Express, Airtable, TinaCMS).
    • Moved the full admissions-to-enrollment flow into the portal: 12 steps mapped to 18 back-office statuses, including a payment plan builder, on-screen signing of the payment agreement and team approval queues.
    • Per-installment Stripe Checkout with webhooks that record each payment against the right installment, replacing manual payment matching by staff.
    • SEO-focused site with a no-code admin hub on TinaCMS for non-technical staff, and a Markdown version of every page for AI agents.
    Live ↗

  5. Forecasting Model in Pharma Product Sales

    • Undergraduate graduation project at Boğaziçi University, supervised by Prof. Aslı Sencer.
    • Team of three; built machine-learning and time-series models in Python to forecast SKU-level sales from six linked industry datasets.

  6. House Price Prediction Platform (Model&Frontend)

    • Decision Support Systems course project: a regression model of King County house prices with a web frontend.
    • Shows which factors drive home prices, as decision support for homebuyers.
    GitHub ↗

  7. Covid-19 Tracker Application

    • Created a user-friendly Covid-19 tracker using React, Redux, Tailwind CSS, and TypeScript.
    • The app provides global and country-specific case data, features an interactive world map, and ensures smooth performance through efficient state management and Docker deployment.
    GitHub ↗

  8. Real Estate Scraper

    • Developed a web scraper that extracts information from a real estate listing website using BeautifulSoup, Pandas, Requests, and Numpy, and saves the extracted data in an excel file.
    GitHub ↗

  9. Course Registration System

    • Developed a web-based course registration system using vanilla JavaScript, PHP, and MySQL.
    GitHub ↗

  10. Find My ID

    • Built the authentication module (username & password reset tool) for a web system using JavaScript, Tailwind CSS, and Next.js.

Skills

Languages
JavaScriptTypeScriptPythonJavaSwiftUISQLNodeJSHTMLCSS
Frameworks/Libraries
ReactNext.jsVueReduxFastAPIStoryBookStrapiCypress
Tools
GitGitHubDockerSentryDatadogCloudflareJiraConfluence
Others
RESTful APIAgileScrumAWSGoogle CloudAzureMySQLPostgreSQL