# Mohammed Efaz Software engineer and remote research assistant at ELITE Research Lab LLC. Interested in cloud engineering, AI inference and GPU programming with CUDA. I'm interested in deploying AI models on cloud infrastructure, optimising inference and writing parallel GPU code with CUDA. I also study agent memory and explore low level systems in Rust, C and C++. I work on Arch Linux with Neovim and a terminal, using Codex, OpenCode and Claude Code for AI assisted development. I'm open to software engineering, cloud engineering and AI inference roles, including GPU programming with CUDA. ## Experience ### ELITE Research Lab LLC · Research Assistant Jun 2026 to present · Remote Headquarters in Queens, New York. - Research Assistant: Sep 2026 to present - Research Intern: Jun 2026 to Aug 2026 I study how AI agents use memory across sessions and build reproducible experiments to evaluate their behaviour. I also contribute to shared research resources and tools for working with AI agents. Tools: ### Genesys · Software Engineer Intern Sep 2024 to Jan 2025 · Budapest Worked on customer journey management workflows, from business logic to tests and bug fixes. - Reached full code coverage in a key journey management area. - Shipped 5+ algorithmic improvements and simplified business logic paths. - Resolved 20+ defects and added 10+ Cypress tests across Angular and Vue modules. Tools: TypeScript, Angular, Vue, Cypress ### Genesys · QA Test Automation Engineer Intern Mar 2024 to Aug 2024 · Budapest Built end to end test coverage for Angular and Vue releases. - Automated 22+ high value end to end scenarios in Playwright and Cypress. - Strengthened CI pipelines with test driven validation for auth, events, and rendering flows. - Supported daily FedRAMP aligned checks in a security sensitive environment. Tools: Playwright, Cypress, TDD, CI/CD ### Eötvös Loránd University · Teaching Assistant Sep 2024 to Jul 2025 · Budapest Guided first year international students through programming fundamentals and lab work. - Mentored 20+ students in coding fundamentals and debugging habits. - Ran lab sessions on programming fundamentals and debugging. Tools: Python, Java, Teaching ## Education Eötvös Loránd University B.Sc. Computer Science · 2022 to 2025 CGPA 4.49/5. Graduated in 2025. ## Tools ### Languages TypeScript, JavaScript, Python, Java, Kotlin, SQL, C#, Rust, C++, Lua ### Interfaces React, Next.js, Vue, Angular, Tailwind, Tauri, Qt, HTML/CSS ### Services FastAPI, Spring Boot, REST API, PostgreSQL, Docker, AWS, Cloudflare, CI/CD ### AI & data PyTorch, TensorFlow, RAG, LangChain, Qdrant, scikit-learn ## Research [MeReader: An Offline, Privacy-Preserving, Progress-Aware AI Assistant for Narrative eBook Reading](https://www.mohammedefaz.com/research/mereader) Authors: Mohammed Efaz, Udo Bub, Itilekha Podder This paper introduces MeReader, a privacy-preserving, offline and progress-aware AI assistant designed to function as a contextual companion for narrative reading on consumer hardware. Unlike conventional Large Language Model (LLM)-based systems that process documents/books online in an ‘all-texts-at-once’ type of way (thereby risking narrative spoilers), MeReader confines both retrieval and generation to the reader’s distinct position in the text. The system integrates a hybrid retrieval architecture, combining vector similarity, BM25, and summary embeddings, to ensure high fidelity to the reader’s current context. We evaluate the system across classical English books using both quantitative comprehension metrics and qualitative LLM-based independent judging. The results identify a quality-latency frontier, showing that mid-sized local models (2b-4b) can achieve comprehension parity with larger counterparts (5b+) while maintaining acceptable response quality. Importantly, the proposed progress-aware boundary mechanism is shown to prevent future content leakage without compromising retrieval quality and correctness. These findings suggest that a localized, privacy-centric approach can augment, rather than replace, the digital reading experience, offering a viable alternative to cloud-dependent services. None of these individual components are new on their own, but no existing system has brought them together in one place. MeReader addresses this gap by combining these capabilities into an offline, constrained assistant that enhances comprehension without compromising continuity or user privacy. ICCCI 2026 · Springer CCIS Citation year: 2027. First online: 18 September 2026. [DOI](https://doi.org/10.1007/978-3-032-37936-8_22) · [Accepted manuscript](https://www.mohammedefaz.com/research/mereader/accepted-manuscript.pdf) ## Projects ### [MeReader](https://www.mohammedefaz.com/project/mereader) Ask about your book with local AI I built an offline ebook assistant that answers questions within your reading progress. Local models use vector and keyword search to find relevant passages. - Implemented progress bound retrieval to prevent spoiler leakage. - Combined vector similarity, BM25, and summary embeddings. - Benchmarked model size tradeoffs for practical local inference. Compared response quality and latency across local models from 2B to 4B. Tools: Python, PyTorch, FastAPI, Vue, Tauri, Qdrant [Source](https://github.com/WhiteHades/mereader) ### [Readest](https://www.mohammedefaz.com/project/readest) I added AI help and library search I built Readest's AI reading assistant with local and cloud models and context limited to your reading progress. My other contributions include full text search, custom shortcuts, paragraph reading, a reading ruler and improvements to highlights and e ink controls. - Implemented the AI reading assistant with local Ollama and cloud models, using progress limited book context. - Added full text library search and fuzzy settings search. - Added paragraph reading, a reading ruler and custom keyboard and mouse shortcuts. - Improved highlight colors, e ink controls, dictionary navigation and remaining time during text to speech. 23 merged contributions covering AI, search and reading controls. Tools: TypeScript, React, Tauri, Rust [My contributions](https://github.com/readest/readest/pulls?q=is%3Apr+author%3AWhiteHades+is%3Aclosed) [Website](https://readest.com/) ### [MeTLDR](https://www.mohammedefaz.com/project/metldr) Summarise email and web pages This Chrome extension summarises Gmail messages, articles and PDFs on your device using Chrome's built in AI or Ollama. You can also ask questions about imported content. - Summarises Gmail messages, articles, and PDFs in the browser. - Uses Chrome's built in AI or a local Ollama model. - Supports questions about imported content. Private, on device summaries for email, articles, and PDFs. Tools: Vue, TypeScript, Chrome Extension, Ollama [Source](https://github.com/WhiteHades/metldr) [Chrome](https://chromewebstore.google.com/detail/metldr-local-ai-gmail-art/kbfdmfgmmibkdnbfdaeganjckfgbfjlh) ### [TheGuideGenie](https://www.mohammedefaz.com/project/theguidegenie) Find and book tours in Budapest I built this Budapest tour booking platform with Stripe payments. Travellers choose available dates, receive confirmations and contact support, while partners manage referrals. - Browse tours, itineraries and available dates before booking. - Pay through Stripe and access the booking confirmation after checkout. - Manage bookings, partner referrals and operations in dedicated workspaces. - Keep support conversations attached to the relevant booking. A deployed booking platform with tour discovery, payments and booking support in one place. Tools: TypeScript, Next.js, Cloudflare Workers, Appwrite, Stripe, Resend [Visit website](https://theguidegenie.com/) ### [shadcn C++](https://www.mohammedefaz.com/project/shadcn-cpp) Build native interfaces with Qt I ported shadcn/ui's components to C++ and Qt, with shared themes and a WebAssembly gallery. Linux is verified, with Windows and macOS checks still pending. - Build native interfaces with buttons, forms, dialogs, navigation, charts and other Qt widgets. - Use shared themes and ordinary Qt parent ownership. - Try real C++ components in the WebAssembly documentation gallery. A released library with a working Linux gallery and browser examples. Windows and macOS verification is still pending. Tools: C++23, Qt 6, CMake, WebAssembly [Source](https://github.com/WhiteHades/shadcn-cpp) [Live gallery & docs](https://whitehades.github.io/shadcn-cpp/) ### [ipynb.nvim](https://www.mohammedefaz.com/project/ipynb-nvim) Edit and run notebooks in Neovim Edit Jupyter cells in Neovim, run kernels and save notebook outputs, including plots in the terminal. I built the notebook integration on an execution runtime adapted from Molten and Magma under GPL version 3. - Edit notebook cells and run them through a Jupyter kernel. - Save notebook outputs and view or enlarge plots in the terminal. - Use notebook language support alongside the normal Neovim editing workflow. A released notebook workflow for Neovim, with installation guidance and terminal output support. Tools: Lua, Rust, Python, Jupyter, Neovim [Source](https://github.com/WhiteHades/ipynb.nvim) [Releases](https://github.com/WhiteHades/ipynb.nvim/releases) ### [melearner](https://www.mohammedefaz.com/project/melearner) Organise local media into courses A Linux app for organising local videos, audio and documents into courses, with saved progress, search and subtitles. It stores your activity on your device and currently installs from source. - Organise local media into courses. - Resume playback and reading with saved progress. - Search course files and browse subtitles and outlines. A local course library with playback, search and saved progress. Tools: C++, Qt, SQLite [Source](https://github.com/WhiteHades/melearner) ### [CodexBar Linux](https://www.mohammedefaz.com/project/codexbar-linux) Check your AI usage limits I maintain the Linux port of Peter Steinberger's CodexBar, with packaged releases for the tray and terminal. It tracks AI usage across accounts with history and alerts. - Check usage from the tray, terminal or JSON commands. - Track history and configure alerts across accounts. Released Linux packages for a native tray and terminal usage monitor. Tools: C, GLib, ncurses, SQLite [Source](https://github.com/WhiteHades/CodexBar-linux) [Linux releases](https://github.com/WhiteHades/CodexBar-linux/releases) ### [MangaStack](https://www.mohammedefaz.com/project/mangastack) Group manga volumes by series I built this Calibre plugin to group manga volumes by series while preserving their archives, covers and metadata. You review proposed changes before applying them, and the released plugin is verified with Calibre 9.4 on Linux. - Group volume records into a visible series container. - Review proposed changes before applying them. - Keep child archives, covers and metadata intact. A released Calibre plugin for browsing manga by series without losing volume records. Tools: Python, Qt, Calibre [Source](https://github.com/WhiteHades/calibre-mangastack) [Plugin releases](https://github.com/WhiteHades/calibre-mangastack/releases) ### [Pokemon TCG](https://www.mohammedefaz.com/project/pokemon-tcg) Browse and filter Pokémon cards A PHP app for browsing Pokémon trading cards and filtering your collection. - Renders card and collection views from server data. - Filters cards for discovery. - Connects browser interactions to backend data. Card browsing, filtering, and collection flows for web users. Tools: PHP, JavaScript, Web [Source](https://github.com/WhiteHades/PokemonTCG) ### [MapMaker](https://www.mohammedefaz.com/project/mapmaker) Build a map and score points Place, rotate and mirror terrain on an 11 by 11 grid in this JavaScript browser game. Mission scores are calculated at the end of each of the four seasons. - Implemented terrain placement on an 11 by 11 grid. - Added controls to rotate and mirror terrain pieces. - Calculated mission scores at the end of each season. Scores terrain placement across four seasons. Tools: JavaScript, HTML, CSS [Source](https://github.com/WhiteHades/MapMaker) ## Contact [GitHub](https://github.com/WhiteHades) · [CV](https://www.mohammedefaz.com/assets/2026_09_18_cv_efaz.pdf) · [LinkedIn](https://www.linkedin.com/in/mohammed-efaz) · [Email](mailto:efaz@mohammedefaz.com)