Tarun Beniwal

Software Engineer

SpeEdLabs

May 2025 - Present

Designed unified multi-database architecture using SQLAlchemy ORM(MySQL) and Beanie ODM(MongoDB), ensuring ACID-compliant consistency and seamless data orchestration for 300K+ active users. Engineered distributed FastAPI services integrating Azure Service Bus and Vimeo APIs with observability and throttling. Strengthened platform security and storage reliability using bcrypt hashing, JWT-based auth, encrypted secrets and Azure Blob with signed URLs, parallel uploads, and integrity checks. Developed fast ETL/ELT pipelines & distributed leaderboard for 270k+ learners, processing 50M+ activity data using vectorized NumPy–Pandas, caching, incremental recomputation & memory-optimized to cut latency by 60%. Built embeddings-driven similarity pipeline across 800K+ questions, enabling large-scale deduplication from content. Developed an AI-assisted PDF Question-Extraction engine using OCR and multi-stage LLM workflows with robust LLMOps safeguards—achieving 75% reduction in manual content effort and production-grade reliability. Speech-to-Text Pipeline scaled a production-grade AI transcription pipeline with Whisper-Large V3 and Gemini, generating high-fidelity transcripts and metadata for Vimeo videos through scalable batching, fallback flows. Engineered scalable ed-tech modules using Vue.js + TypeScript (Pinia, Vue Router), integrating LaTeX/HTML editors and RBAC workflows for secure, multi-tenant user experiences. Modernized frontend by migrating from Bootstrap/Element Plus to Tailwind, Ionic, and Capacitor, improving mobile responsiveness 40% and enabling seamless cross-platform delivery (web, Android, iOS). Designed reusable, event-driven UI systems using Ionic components and Swiper.js, optimizing slides, forms, modals, and navigation, reducing interaction latency 30% and improving engagement 20%.