Agent

The agent can understand and process text, images, PDFs, audio, and YouTube links, ask clarification questions when needed, and execute tasks with structured, explainable outputs.

What This Agent Can Do

Understand Multimodal Inputs

• Text
• Image screenshots (OCR via Gemini Vision)
• PDF files (digital + scanned OCR fallback)
• Audio (Whisper STT)
• YouTube links (transcript extraction)

High Level Architecture

LangGraph Workflow

Demo Video

Click the image below to watch the full demo

Tech Stack

• FastAPI (backend API)
• LangGraph (agent orchestration)
• Gemini 2.5 Flash (LLM + Vision)
• Gemini OCR (for images + scanned PDFs)
• pdfplumber + pdf2image (fallback extraction)
• Whisper (audio transcription)
• React (frontend)

---

Running Locally

Install dependencies

cd agent
pip install -r requirements.txt

Start backend

uvicorn main:app --reload

Start frontend

cd frontend
npm install
npm run dev