The Problem
Long PDFs, contracts, and reports bury the answer people actually need. Teams end up Ctrl+F-ing through dozens of pages, or worse, trusting a summary with no way to verify where it came from. The same gap shows up in hiring: an HR team with a stack of resumes has no fast way to ask "who has 3+ years of React experience" and get a grounded, page-accurate answer across every file at once.
What We Actually Did
Built a full-stack Retrieval-Augmented Generation pipeline: documents are chunked and embedded, stored in a vector index, and retrieved by semantic similarity at query time. Every answer the LLM generates is grounded in the retrieved chunks and returned with a page-level citation, so the source is always one click away. The entire app, frontend and serverless backend, ships as a single Next.js project deployed on Vercel's free tier, with no separate server to provision or maintain.
The Solution
Ragdesk turns any uploaded PDF into a conversation. Ask a question in plain language and get an answer backed by exact page citations, so nothing has to be taken on faith. The same engine scales to multi-document use cases like HR teams comparing several resumes side by side, or legal and compliance teams querying long policy documents, without losing traceability back to the source.
System Infrastructure
Powered by an LLM + Vector Search retrieval layer running entirely inside Next.js API routes. Embeddings are generated on upload, indexed for similarity search, and queried live on every question, with the model prompted to answer strictly from retrieved context. Deployed as a single app on Vercel's free tier, keeping the whole stack serverless and cost-free to host.