🦖 RAWR Agent – Read And Write Report

• A starter workflow template for NVIDIA AIQ Toolkit by Jeremy Kesten (https://linktr.ee/JeremyKplatform ) 

• Produced for and submitted to the Nvidia AgentIQ Hackathon 2025 - https://developer.nvidia.com/agentiq-hackathon

TL;DR Access the Nvidia AgentIQ chat UI locally, to prompt an agent to look at an image (URL or local file) → the first agent tool asks your preferred vision enabled LLM or NimOCR to extract structured JSON → a second agent tool function converts the JSON into Report Creator (https://github.com/darenr/report_creator) templates → instantly get a polished, single‑file HTML report.

Recommended Viewing on Nvidia AgentIQ - https://www.youtube.com/watch?v=H65OluZaiZQ&t=153s

---

🌟 What’s inside?

Feature	Details
Image → JSON	Uses a vision enabled LLM (default: build.nvidia.com, easily swappable for OpenAI, etc.) to “look” at an image and answer with JSON. Also includes example scripts for running a PaddleOCR NIM docker container, for more sensitive applications, where an external LLM is not possible. Resizes large images before scanning to save tokens. Extended formats for PaddleOCR NIM - can scan JPEG, PNG, GIF, TIFF, WEBP, BMP
Template‑driven reports	Report Creator renders the JSON into eye‑catching HTML via the templates in report_templates/.
One‑call flexibility	For multiple report types, you can pick any custom pre-setup template just by passing its name in a single function‑call parameter.
Batteries included	Works as a complete AIQ workflow, built to be a scaffolding so you can extend the logic between a basic example of vision llm “read” and report “write.” For instance, add an MCP extend analysis of the vision data, before writing the report.

---

🦕 Getting Started

1. Download AIQToolkit

   mkdir -p ~/<your‑AIQtoolkit‑dir>
   git clone https://github.com/NVIDIA/AIQToolkit
   #  for local dev we install in WSL in ~/aiqtoolkit 

2. Clone or fork this AgentIQ Workflow

   cd <your‑AIQtoolkit‑dir>
   mkdir -p ~/<your‑AIQtoolkit‑dir>/workflows
   cd workflows
   git clone https://github.com/jeremykpark/rawr_agent.git

Assumes you already have Python 3.9 + and uv or pip handy.

3. Install AIQToolkit

   Create a new .venv environment and Install AIQTOOLKIT
   #  follow the official AIQToolkit setup guide :
   #  https://docs.nvidia.com/aiqtoolkit/latest/quick-start/installing.html
   

4. Register this workflow with AIQ

   cd ~/<your‑AIQtoolkit‑dir>/workflows/rawr_agent
   uv pip install -e .
   
   

5. Launch the workflow server from the workflow root dir

    cd ~/<your‑AIQtoolkit‑dir>/workflows/rawr_agent
   aiq serve --config_file configs/config.yml

6. Launch the AIQ UI in a separate terminal

   cd ~/<your‑AIQtoolkit‑dir>/external/aiqtoolkit-opensource-ui
   npm run dev
   
   # follow the official AIQ UI setup guide first - NPM version v18.17.0 or new required:
   # Ref: https://docs.nvidia.com/aiqtoolkit/latest/quick-start/launching-ui.html

6.5. ** Optional ** Start NimOCR server in a seperate terminal

cd nim_ocr
start_ocr_nim.sh

7. Open your browser at http://localhost:3000
   Start chatting - test with one of the prompts in README_prompt_suggestions.md; watch the server terminals for logs/errors.

8. Go RAWR in the jungle
   Use the chat prompt to point the agent at any image and enjoy the auto‑generated reports in report_exports/ (self‑contained interactive HTML). There are suggested prompts in README_prompt_suggestions.md to get you started.

---

🖼️ Sample Flyer Included

The default setup is to read this included sample file and create a report on the data inside.

| | |

---

🧩 Template System

Create new templates in report_templates/. Set them up in rawr_report_template.py to call them by name. Generated reports are saved to /report_exports as an HTML file.

Click here to see "Kitchen Sink" an example of some of the widgets you can access for your own reports: Kitchen Sink Widget Sample

Reports are powered by Report Creator by Daren Race.

---

📽️ Walkthrough Video

---

🏗️ Build Your Own RAWR

This repo is deliberately minimal and is to be used as scaffolding for your own projects. Insert your own logic between read (LLM vision) and write (HTML report):

graph LR
  A(Image) -->|LLM Vision| B(JSON)
  B -->|Your Code 🤖| C(Enhanced JSON or extra data)
  C -->|Report Creator| D(Beautiful HTML)

Loading

---

🌋 Contributing / Issues

It’s a public repo—PRs & issues are welcome! Let’s make this starter template an even louder RAWR.

“Ah, now eventually you do plan to have dinosaurs on your, on your dinosaur tour, right? Hello?” — Dr. Ian Malcolm

---

📄 License

Distributed under the MIT License. See LICENSE for more information.