AskkkDoc: ask questions about your own documents with AI

AskkkDoc lets you upload your own PDFs, Word documents and images, then ask questions and get answers based on those files. Each user gets 3 free documents.

AskkkDoc is my own project. I designed, built and shipped it myself, and it is not client work.

At a glance

Year
2024
Status
Live demo (may go offline)
Role
Design and development, all done by me.
Stack
Next.js, TypeScript, Tailwind, Shadcn UI, Next Auth, Prisma, MySQL on PlanetScale, AWS S3, Pinecone, OpenAI, LangChain, Unstructured, Vercel AI SDK, Docker (local database)
Links

What I built

The upload and file processing pipeline, the question and answer feature, sign-in and the database. It is a monorepo with the backend in Next.js API routes. I never built a higher document limit.

My role: Design and development, all done by me.

How it works

  1. You upload a PDF, a Word document or an image.
  2. The file is stored in AWS S3.
  3. LangChain with the Unstructured loader turns the file into text.
  4. The text is converted into vectors and stored in Pinecone.
  5. When you ask a question, OpenAI writes the answer using those vectors. Your user data lives in MySQL.

What I learned

Handling PDFs, Word files and images taught me to treat file processing as its own pipeline: store the file, turn it into text, turn the text into vectors, then answer from those vectors. Keeping each step separate made failures easy to find.

Want something like this built?

If you need this kind of work, see my page on MVP development for founders. You can also see all my projects.