PaperShredder AI

Intro section (Home Page)
Result section (Output Page)
Wallet section(Credentials)
Setting section (Profile Page)

🔥 Inspiration

It can be exhausting to read scientific papers. Peer review is slow, bias is easy to slip through, and bad research methodology often goes unchecked. We wanted to build an AI agent that does more than “summarize” research. That's how PaperShredder AI was born, the ultimate skeptical research assistant.

🧠 What it does

PaperShredder AI reads and deconstructs any PDF file to find:

Credibility Scoring: Evaluates the robustness of the paper.
Claim Extraction: Identify the boldest or riskiest claims that the authors make.
Contradiction Matrix: Cross-references those claims against data to find contradictions.
The "Savage Peer Review": Enables you to hear a raw, brutal critique of the flaws of the paper as an entertaining podcast clip.

⚙️ How we built it

We designed a full-stack web app in React, Vite, and TypeScript which is fully supported by three powerful sponsor APIs.

Google Gemini API: This is our AI. We use Gemini Multimodal Vision to read the PDF file, extract claims, detect fallacies, and write the savage peer review.
Snowflake API: This is our secure data backbone. We developed a robust authentication system and store all user profiles, claims, and analysis history in Snowflake tables.
ElevenLabs: This is how we bring the AI to life. We created an interactive widget called AI Voice Reviewer that reads the savage peer reviews with incredibly realistic and expressive audio.

🧩 Challenges

The biggest challenge was putting it all together: pulling content from PDFs, passing it to Gemini in strict JSON form, storing the data in Snowflake, sending it to ElevenLabs for real time audio, all in one seamless step. We also had to make sure we were getting the prompting just right so that the AI sounded critical and intelligent, not mean.

🏆 What we learned

We learned the power of keeping data in a structured fashion (Snowflake), reasoning (Gemini) and being able to express ourselves with human-like expressiveness (ElevenLabs). AI is not only for summarizing: When it is properly orchestrated, it can verify factual content.

🚧 What’s next

We plan to expand our Snowflake database to store millions of verified claims, so that PaperShredder can cross-reference new papers to a large historical contradiction matrix immediately.

Built With

css
elevenlabs
gemini
radix
react
snowflake
typescript
vite

Submitted to

AI Hackfest

Created by

I acted as the full-stack lead for this project. I was responsible for the core frontend architecture using React Router v7 and Vite. I also orchestrated the Google Gemini API integration, ensuring the multimodal vision model could accurately parse the dense PDFs and extract the structured claims used in our contradiction matrix.

Rushikesh Bobade
I worked heavily on the backend integration and data persistence. My main focus was integrating the Snowflake REST APIs to ensure our user authentication, session state, and historical paper analyses were securely saved. It was a great challenge mapping the complex AI output into structured database tables!

Afroz Khan
I focused on the UI/UX design and the audio integration. I integrated the ElevenLabs API to create the "AI Voice Reviewer" widget, handling the text-to-speech audio streaming so the Savage Peer Review could be read aloud. I also managed the routing architecture and frontend documentation.

Shravan Navale
I worked on the frontend build systems and state management. I configured the Vite build pipeline and helped build the global state hooks that pass the AI analysis data (claims, credibility scores, contradictions) between the processing pages and the final results dashboard.

Chaitrali Tikar