Using Agentic RAG for Research - Demonstration of NotebookLM

Using Agentic RAG for Deep Research
Dr.Rathishchandra R. Gatti
Professor & Dean –Research
gattirathish@gmail.com
Research Seminar 01
Topic : Using Agentic RAG for Deep Research RIG: Agentic AI , Robotics and AR/VR
Resource Person: Dr.Rathishchandra R Gatti
Time: 03-Dec-2025, 4:00-4:30 PM
Venue: For Internal participants: Ground floor Seminar Hall
For external participants: Google Meet joining info - ttps://meet.google.com/nrm-zdfu-xfdOr dial: (US) +1 402-523-0291 PIN: 991 383 769#
From Research Interest Group :
Agentic AI , Robotics and AR/VR
Dr.Rathishchandra R. Gatti , gattirathish@gmail.com . 1

Agenda
1. What is RAG ?
2. Demo –NotebookLM

What is RAG? (Retrieval Augmented Generation)
Source : https://www.geeksforgeeks.org/nlp/what-is-retrieval-augmented-generation-rag/
Retrieval-Augmented
Generation (RAG) is an
advanced AI framework that
combines information retrieval
with text generation models
like GPT to produce more
accurate and up-to-date
responses. Instead of relying
only on pre-trained data like
traditional language models,
RAG fetches relevant
documents from an external
knowledge source before
generating an answer.

Importance of RAG
1.Access to Updated Knowledge: LLMs are trained on fixed datasets but RAG allows them to fetch fresh
and real time information from external sources.
2.Improved Accuracy: It reduces hallucinations in LLMs and makes answers more factually correct.
3.Domain Specific Expertise: It lets us use specialized datasets like medical records and legal documents
to get expert-level responses without retraining the model.
4.Cost Efficiency: Instead of retraining massive LLMs with new data, we simply update the external
knowledge base hence saving time and resources.
5.Personalization: RAG can retrieve user specific information like past interactions or personal data to
provide more tailored and relevant responses

Working of RAG
The system first searches external sources for relevant information based on the user’s query instead of
relying only on existing training data.

1.Creating External Data: External data from APIs, databases or documents is chunked, converted into
embeddings and stored in a vector database to build a knowledge library.
2.Retrieving Relevant Information: User queries are converted into vectors and matched against stored
embeddings to fetch the most relevant data ensuring accurate responses.
3.Augmenting the LLM Prompt: Retrieved content is added to the user’s query giving the LLM extra context
to work with.
4.Answer Generation: LLM uses both the query and retrieved data to generate a factually accurate, context
aware response.
5.Keeping Data Updated: External data and embeddings are refreshed regularly in real time or scheduled
so the system always retrieves latest information.

What Problems does RAG solve?
Some the problems that RAG solves are:
1.Hallucinations: Traditional generative models can produce incorrect information. RAG reduces this risk by
retrieving verified, external data to ground responses in factual knowledge.
2.Outdated Information: Static models rely on training data that may become outdated. It dynamically retrieves
latest information ensuring relevance and accuracy in real time.
3.Contextual Relevance: Generative models often struggle with maintaining context in complex or multi turn
conversations. RAG retrieves relevant documents to enrich the context improving coherence and relevance.
4.Domain Specific Knowledge: Generic models may lack expertise in specialized fields. It integrates domain
specific external knowledge for tailored and precise responses.
5.Cost and Efficiency: Fine tuning large models for specific tasks is expensive. It eliminates the need for retraining
by dynamically retrieving relevant data reducing costs and computational load.
6.Scalability Across Domains: It is adaptable to diverse industries from healthcare to finance without extensive
retraining making it highly scalable.

Challenges
Despite its advantages, RAG faces several challenges:
1.Complexity: Combining retrieval and generation adds complexity to the model requires careful tuning
and optimization to ensure both components work seamlessly together.
2.Latency: The retrieval step can introduce latency making it challenging to deploy RAG models in real
time applications.
3.Quality of Retrieval: The overall performance heavily depends on the quality of the retrieved
documents. Poor retrieval can lead to suboptimal generation, undermining the model’s effectiveness.
4.Bias and Fairness: It can inherit biases present in the training data or retrieved documents,
necessitating ongoing efforts to ensure fairness and mitigate biases.

RAG Applications
Here are some examples to illustrate the applications of RAG we discussed earlier:
1.Question-Answering Systems: It enables chatbots or virtual assistants to pull information from
a knowledge base or documents and generate accurate, context aware answers.
2.Content Creation and Summarization: It can gather information from multiple sources and
generate concise, simplified summaries or articles.
3.Conversational Agents and Chatbots: It enhances chatbots by grounding their responses in
reliable data making interactions more informative and personalized.
4.Information Retrieval: Goes beyond traditional search by retrieving documents and generating
meaningful summaries of their content.
5.Educational Tools and Resources: Provides students with explanations, diagrams or
multimedia references tailored to their queries.

NotebookLM
NotebookLM is effectively a specialized RAG (Retrieval-Augmented
Generation) system.
In fact, Google engineers often refer to
its core mechanism as "source-
grounding" or "RAG-Locked," meaning
it is strictly designed to retrieve
information only from the documents
you upload, rather than drawing from
the AI's general training data to answer
your questions.

NotebookLM
How it is like a Standard RAG pipeline?
It follows the exact same architectural steps as any RAG application:
1. Ingestion: You upload sources (PDFs, Google Docs, text files).
2. Indexing: NotebookLM "reads" these files and creates a searchable index (likely using
vector embeddings).
3. Retrieval: When you ask a question, it searches your documents for the most relevant
chunks of text.
4. Generation: It sends those relevant chunks to the LLM (Gemini 1.5 Pro) along with your
question to generate an answer.
5. Citations: Like a good RAG system, it provides "footnotes" (citations) showing exactly
which part of the document was used to generate the answer.

NotebookLM
How is it different from a standard RAG pipeline?
While it is a RAG, it solves two of the biggest problems with "standard" RAG systems: Context Window and Hallucination.
Feature Standard RAG Pipeline NotebookLM (Gemini 1.5 Pro)
Context Size
Typically retrieves small "snippets" (e.g., 5-
10 paragraphs) because the AI's memory is
small.
Uses a massive context window (up to 2 million
tokens). It can often "retrieve" and "read" your
entire document at once rather than just tiny pieces.
Reasoning
Often struggles to connect dots between
distant parts of a document because it only
sees small fragments.
Because it can "see" much more text at once, it is
significantly better at holistic reasoning (e.g.,
"Summarize the themes across these 5 different
papers").
Knowledge
Source
Often "leaks" outside knowledge (e.g., if you
ask about a specific law, it might mix up your
document's text with what it learned from
the internet).
It uses a "Source-Grounding" technique that
aggressively restricts the model to only use the info
in your documents.

3. Why researchers must go with Notebook LM over other RAGs
Trust: If NotebookLM can't find the answer in your documents, it is designed
to say "I don't know" rather than making something up(Hallucinations).
Complexity: You don't need to set up a vector database (like Pinecone or
Chroma) or write Python code to chunk your documents. Google handles the
entire RAG pipeline automatically. This is useful for those researchers who
donot like/know to code but need a RAG pipeline to focus point their
research to restricted documents.
NotebookLM

NotebookLM Demo

NotebookLM Demo
Click “Create Notebook LM”

NotebookLM Demo
Click “upload sources”
and brows to your
folder where you have
your downloaded
research papers from
science direct
Note : Before this step , You MUST have already downloaded
the research papers from Science Direct OR IEEE Explore OR
any other Research Repository and saved them in your local
drive.

NotebookLM Demo
This is the NotebookLM UI that will appear once you have
finished the upload
This is the Source
window wherein all
the resources that
you have uploaded
can be seen and
managed.
This is the chat
window wherein
you can ask any
questions related
to the research
papers that you
have downloaded
Enter any question
in this chat
question window
See the Answers in
this Answer
window
This is the studio
window wherein
you can create
multimodal
research outputs
(reports, videos,
podcasts )
1
2

NotebookLM Demo
Ask specific questions laser
focused on your research
topic
It displays the number of
resources
It will also display the most
likely questions that you
can ask to get your R&D in
the right direction.

NotebookLM Demo
Ask specific questions laser
focused on your research
topic
Click on the intext citation
Inspecting the answer ..
This window pops up
explaining about the
reference of the citation
you have clicked.
It also provides the
reference of the
document being
uploaded
1
2
3
This is very likely of RAG system ,wherein you get answers
from the domain documents that you have uploaded and
not “beating around the bush” answers.

NotebookLM Demo
If you are happy with the answer and you like
to store it , click “Save to note” .
Anything you have created or saved will
appear always in this studio window.

NotebookLM Demo
This is also useful for creating tables of
comparisons that we frequently enlist in our
systematic reviews

NotebookLM Demo
Exploring the Studio Window
These are studio widgets from which you can
create videos, quizzes and other outputs from
the documents that you have uploaded.
From research perspective , Infographic widget
,Mind maps,SlideDeck and Reports are very
useful.
All your creations from the chats and
the studio widgets will be stored in the
Studio

NotebookLM Demo
Click on the item
you have created
The item
expands in the
studio window
Navigating the Studio Window
The beauty is that you can convert this
creation as a source for the next chat .
This makes our research recursive!
1 2 3

NotebookLM Demo
Click on the Infographic widget in the studio window and do the settings
according to your requirements.
Describe clearly the prompt for the infographic.
You can use your prompting skills. I am a layman
in this prompting business () and have used
very simple prompt.
1
Exploring the Widgets :
2

NotebookLM Demo
Wait for your creation . Grab a cup of tea or any
other drink ;) and just wait.
You can also try doing other creations such as
reports, mindmaps.
If you donot have time , you can even exit the
Application. It will contrinue to generate and the
creations will be in the studio , when you visit this
notebook tomorrow.

NotebookLM Demo
This is the infographic that it has created. Closer to a graphical abstract
The infographic is pretty generic and is useful as a primer source in the
introduction sections of any research paper on edge robotics and SLMs.
If you are using any part of the infographic in the research paper,
DONOT forget to MENTION the AI tool used .
This is not only ethical but ensures transparency and trust among the
authors reviewers and publishers.
SUPER IMPORTANT (Especially for new scholars)

Thank You
For any queries and collaboration ,contact
Dr. Rathishchandra R Gatti
gattirathish@gmail.com
Rathishchandra.mech@sahyadri.edu.in

Using Agentic RAG for Research - Demonstration of NotebookLM

More Related Content

Similar to Using Agentic RAG for Research - Demonstration of NotebookLM

More from Rathish Chandra Gatti,Ph.D

Recently uploaded