NeroEd

NeroEd — a full-stack AI study assistant I built to turn any uploaded document into an interactive learning workspace.

The idea was simple: students shouldn't have to leave their study material to get help understanding it. So I built a system where the document itself becomes the source of truth for everything.

🔍 Here's what NeroEd does:

→ Upload a PDF, DOCX, or TXT file

→ Chat with it — answers grounded strictly in your document

→ Generate flashcards, quizzes, and structured notes automatically

→ Extract formulas and render them in LaTeX

→ Switch between study modes in a clean, tabbed interface

⚙️ Under the hood:

The backend (FastAPI + Ollama) runs a multi-stage retrieval pipeline — hybrid semantic + BM25 search, followed by cross-encoder reranking — to surface the most relevant passages before generating any response. No hallucination, no generic answers.

The frontend (Next.js) streams responses in real time and renders the full study workspace with Markdown and KaTeX support.

Building NeroEd pushed me to think seriously about RAG architecture, document chunking strategies, and what it actually means to design an AI tool around a specific user workflow — not just a chat box.

Happy to answer questions about the architecture or the decisions behind it. 👇

hashtag#AI hashtag#FullStack hashtag#RAG hashtag#NextJS hashtag#FastAPI hashtag#MachineLearning hashtag#EdTech hashtag#BuildInPublic