Build Multimodal AI Agents for Document and Video Intelligence using NVIDIA Nemotron Nano 2 VL

[NavuluriBalaji]

Build Multimodal AI Agents for Document and Video Intelligence using NVIDIA Nemotron Nano 2 VL

Abstract of the talk/workshop
This talk explores the Architecture and capabilities of Nemotron Nano 2 VL, NVIDIA's compact powerful vision-language model, in building intelligent agents that can process and understand both documents and videos. Participants will learn how to use this multimodal AI model to create applications that can extract insights from complex documents, analyze video content, and generate contextual responses. The session will cover practical implementation strategies, including setting up the model, and integrating with existing AI pipelines. Real-world use cases will demonstrate how these agents can transform document processing workflows and enable advanced video intelligence in various industries. Attendees will gain hands-on insights into building efficient AI solutions that balance performance with computational resources.

Category of the talk/workshop
Data Science, Machine Learning, and AI

Duration (including Q&A)
40 minutes (30 minutes presentation + 10 minutes Q&A)

Level of Audience
Intermediate

Speaker Bio
Name: Navuluri Balaji
Company: AVK Tech Solutions
Position: Associate Software Developer
Email: navuluribalaji03@gmail.com
Years of Experience: 1
Portfolio: https://linktr.ee/BalajiNavuluri

Prerequisites(if any)
Basic understanding of AI/ML concepts
Familiarity with Python programming
Basic knowledge of neural networks and transformers (helpful but not required)
No specific software setup required as the talk will focus on concepts and implementation approaches

[kalyan678]

@NavuluriBalaji - Thanks for submitting the proposal! The topic on building multimodal AI agents with NVIDIA Nemotron Nano 2 VL sounds like a great fit for our upcoming HydPy meetup.
To help us structure the agenda and ensure attendees get a hands-on takeaway, could you please share a brief outline of your session flow for example, the key sections you plan to cover, and any demos or code walkthroughs (if available)? This will help us better understand the session’s structure and plan accordingly.

[NavuluriBalaji]

Hi Team, Thank you for the update and for considering my proposal. As requested, I’m sharing a brief outline of the session flow for the talk *“Build Multimodal AI Agents for Document and Video Intelligence using NVIDIA Nemotron Nano 2 VL.”* Session Flow Outline

*1. Introduction*Quick intro and context on why multimodal intelligence matters. Shift from text-only LLMs to vision-language models (VLMs). *2. Vision-Language Models (VLMs) Overview*What VLMs are and how they differ from traditional LLMs. Key architecture differences and reasoning capabilities. *3. Nemotron Nano 2 VL Overview*Model capabilities, architecture components, and supported modalities. Why Nemotron Nano 2 VL is suited for document + video understanding. *4. Architecture Decoding & Inference Analysis*End-to-end workflow: image/video/document processing → embeddings → multimodal fusion → response. Comparison with other VLMs like LLaVA, Qwen-VL, etc. *5. Hands-on Demo / Codelab (Google Colab)*Setting up and running nemotron-nano-12b-v2-vl. Inference on images, documents, and video frames. Building a simple multimodal agent for contextual Q&A and insight extraction. Practical examples showcasing document intelligence and video analysis.

7. Q&A Thanks & Regards, *Navuluri Balaji*

…

On Wed, 12 Nov 2025 at 23:27, Runtime Terror ***@***.***> wrote: *kalyan678* left a comment (HydPy/HydPy-meetups#149) <#149 (comment)> @NavuluriBalaji <https://github.com/NavuluriBalaji> - Thanks for submitting the proposal! The topic on building multimodal AI agents with NVIDIA Nemotron Nano 2 VL sounds like a great fit for our upcoming HydPy meetup. To help us structure the agenda and ensure attendees get a hands-on takeaway, could you please share a brief outline of your session flow for example, the key sections you plan to cover, and any demos or code walkthroughs (if available)? This will help us better understand the session’s structure and plan accordingly. — Reply to this email directly, view it on GitHub <#149 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A4EFRIX7N26JFHV73F43E4L34NYJRAVCNFSM6AAAAACK2ZYXS2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKMRTGE4DOMRRGQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[kalyan678]

Hi Team,

Thank you for the update and for considering my proposal. As requested, I’m
sharing a brief outline of the session flow for the talk

“Build Multimodal AI Agents for Document and Video Intelligence using
NVIDIA Nemotron Nano 2 VL.”
Session Flow Outline

1. IntroductionQuick intro and context on why multimodal intelligence
matters.
Shift from text-only LLMs to vision-language models (VLMs).

2. Vision-Language Models (VLMs) OverviewWhat VLMs are and how they
differ from traditional LLMs.
Key architecture differences and reasoning capabilities.

3. Nemotron Nano 2 VL OverviewModel capabilities, architecture
components, and supported modalities.
Why Nemotron Nano 2 VL is suited for document + video understanding.

4. Architecture Decoding & Inference AnalysisEnd-to-end workflow:
image/video/document processing → embeddings → multimodal fusion → response.
Comparison with other VLMs like LLaVA, Qwen-VL, etc.

*5. Hands-on Demo / Codelab (Google Colab)*Setting up and running
nemotron-nano-12b-v2-vl.
Inference on images, documents, and video frames.
Building a simple multimodal agent for contextual Q&A and insight
extraction.
Practical examples showcasing document intelligence and video analysis.
7. Q&A

Thanks & Regards,
Navuluri Balaji
…

@NavuluriBalaji - Thanks for making the changes. Are you available to present this in person on Dec, 6th 2025?

[NavuluriBalaji]

Hi Team, Hope you're doing well Yes, I will be available to present *in person on December 6th, 2025*. Please let me know if there are any additional details needed from my side. Looking forward to the session! Thanks & Regards, *Navuluri Balaji*

[bpkapkar]

Hi @NavuluriBalaji ,

Thank you for the confirmation.
Request you to please share your photo so we can get the event flyer ready.
We will be allocating 30 minutes for your session: 20–25 minutes for the talk and 5 minutes for Q&A. Kindly plan the session accordingly so it creates a high impact on the audience.

Thanks & Regards,
Bhushan Kapkar

[NavuluriBalaji]

Hi @bpkapkar,

Thank you for the update.
I’ve shared my photo via Google Drive — please find it attached below
https://drive.google.com/file/d/1_kEVsxGdhVOZ-lZ1bImHRPAelPr-5LFZ/view?usp=sharing
I’ll structure the session to fit within the allocated 30 minutes and ensure it delivers strong value.

Thanks & Regards,
Navuluri Balaji

[bpkapkar]

Hello @NavuluriBalaji ,

Due to some internal challenges, we will need to postpone your talk and plan to schedule it for a future meetup. For the upcoming NVIDIA collaboration event, we will not be able to host your session. We will keep you posted on our upcoming meetup plans.

Thank you for your understanding.

[NavuluriBalaji]

Hi @bpkapkar

Thank you for letting me know. I’m happy to have my session considered for a future meetup soon.