Annual-Report-AI

Inspiration

Growing up, a kid's nightmare was to be assigned homework that involved reading tons and tons of information. Manually finding key information is slow, boring, and time consuming—so we built a tool to fix that.

What it does

Upload any PDF and ask it questions. Our app uses embeddings and LLMs to understand the document and give you accurate, context-aware answers instantly.

How we built it

We used FastAPI for the backend and served a simple HTML/CSS frontend. PDFs are processed with pdfminer, chunked using Tesseract, embedded with OpenAI, and queried via a RetrievalQA chain.

Challenges we ran into

• Managing OpenAI API errors like invalid keys and rate limits.
  
• Integrating the Tesseract module with community packages.
  
• Handling large PDFs and having the API successfully access them.

Accomplishments that we're proud of

• Built a functional PDF Q&A system in under 48 hours.
  
• Clean, easy-to-use UI and backend.
  
• Scalable design that works on any kind of PDF.

What we learned

• How to combine vector databases with LLMs effectively.
  
• Deployment and debugging of FastAPI apps.
  
• API rate handling, Tesseract practices, and collaborative GitHub workflows.

What's next for Annual-Report-AI

• Support for other LLMs like Gemini or Claude.
  
• OCR support for scanned/image-based PDFs.
  
• User accounts with PDF history.
  
• Deploying a public version accessible by anyone.

Built With

• css
• fastapi
• html
• javascript
• langchain
• pdfminersix
• python
• tiktoken
• vscode