OpenScholar Live Academic Research Engine

Built a serverless academic research engine that solves the information-lag problem inherent in static RAG systems by querying the live arXiv API in real time rather than relying on pre-indexed corpora.

Downloaded papers are parsed, chunked, embedded on the fly using sentence transformers, and indexed into Qdrant Cloud with active payload indexing keyed to individual paper IDs for isolated per-session workspaces.

The synthesis layer uses Groq as the primary high-throughput inference engine and automatically falls back to Google Gemini under rate-limit conditions, ensuring uninterrupted operation at zero cost.

Frontend is a dual-pane React application built with Vite and Tailwind CSS v4, exposing an exploratory search tracking panel alongside a deeply synthesized literature review output.

Backend FastAPI service containerized and deployed to Hugging Face Spaces; frontend deployed to Vercel with automated production rollouts via GitHub integration.