A PDF chat template that leans hard on hybrid search

What it does PDFToChat is a template RAG application that lets users upload PDFs and ask questions about their contents. It stores PDFs via Bytescale, chunks and indexes them in Chroma Cloud using hybrid dense-plus-sparse retrieval, and generates answers with Mixtral through Together AI. The frontend is Next.js with Clerk handling authentication and Vercel hosting the lot.

The interesting bit The project leans on Chroma Cloud to run both Qwen dense embeddings and SPLADE sparse embeddings, then fuses the results with Reciprocal Rank Fusion at query time. That is a lot of search sophistication for a boilerplate, and it means you do not need to run a local embedding model.

Key highlights

• Hybrid retrieval via Chroma Cloud (Qwen + SPLADE with RRF) with automatic per-document collections
• LLM inference through Mixtral on Together AI, orchestrated with LangChain.js
• Auth, storage, hosting, and database are all external SaaS (Clerk, Bytescale, Vercel/Postgres)
• Future task list is endearingly honest: no delete button, no chat history, no API route auth, and no benchmarks yet

Caveats

• The future tasks list is longer than the feature list; deleting PDFs, saving chats, protecting API routes, and benchmarking retrieval are all explicitly marked as not yet done
• Running your own instance requires accounts and API keys for Together AI, Chroma Cloud, Bytescale, Clerk, and Vercel Postgres
• The README warns that Together AI’s free tier rate limits can block requests until you add a credit card

Verdict Good for developers who want to see how a modern, fully-outsourced RAG stack fits together in Next.js. Skip it if you are looking for a production-ready, self-contained document chat system.