Index
A
Activeloop
URL 40
Activeloop Deep Lake 32, 33
adaptive RAG 118-120
selection system 125
advanced RAG 4, 21
index-based search 24
vector search 22
Agricultural Marketing Service (AMS) 203
AI-generated video dataset 263
diffusion transformer model video dataset, analyzing 266
diffusion transformer. working 264-266
Amazon Web Services (AWS) 146
Apollo program
reference link 41
augmented generation, RAG pipeline 50, 51
augmented input 53, 54
input and query retrieval 51-53
B
bag-of-words (BoW) 221
Bank Customer Churn dataset
collecting 146-151
environment, installing for Kaggle 148, 149
exploratory data analysis 151-153
ML model, training 154
preparing 146-148
C
Chroma 214, 215
Chroma collection
completions, embedding 220, 221
completions, storing 220, 221
data, embedding 218, 219
data, upserting 218, 219
embeddings, displaying 221
model, selecting 219
content generation 132-134
cosine similarity
implementing, to measure similarity between user input and generative AI model's output 56-58
D
data embedding and storage, RAG pipeline 44, 45
batch of prepared documents, retrieving 45, 46
data, adding to vector store 47, 48
embedding function 47
vector store, creating 46
vector store information 48-50
vector store, verifying for existence 46
data embeddings 33
data, for upsertion
preparing 193, 194
dataset
downloading 239
preparing, for fine-tuning 239-242
visualizing 240
Davies-Bouldin index 156
Deep Lake API
reference link 48
Deep Lake vector store
creating 194
populating 194
diffusion transformer model video dataset
analyzing 266
thumbnails and videos, displaying 270-272
video download and display functions 266-268
video file 268-270
documents
collecting 188
preparing 188
dynamic RAG
applications 210
architecture 210-212
collection, deleting 230, 231
collection, querying 222-225
dataset, downloading 216, 217
dataset, preparing 216, 217
environment, installing 212
prompt 225
prompt response 227
query result, retrieving 227
session time, activating 215, 216
total session time 231
using, with Llama 227-230
dynamic RAG environment installation
Chroma 214, 215
Hugging Face 213, 214
E
embedding models, OpenAI
reference link 47
embeddings 32
entry-level advanced RAG
coding 9
entry-level modular RAG
coding 9
entry-level naïve RAG
coding 9
environment
installing 238, 239
environment setup, RAG pipeline 36
authentication process 39, 40
components, in installation process 36, 37
drive, mounting 37
installation packages 36
libraries 36
requisites, installing 39
subprocess, creating to download files from GitHub 37, 38
evaluator 8, 134
cosine similarity score 134
human-expert evaluation 137-140
human feedback 9
human user rating 135-137
metrics 9
response time 134
F
fine-tuned OpenAI model
using 246-248
fine-tuning
dataset, preparing for 239-242
versus RAG 4
fine-tuning documentation, OpenAI
reference link 248
fine-tuning static RAG data
architecture 236, 237
foundations and basic implementation
data, setting up with list of documents 13
environment, installing 10, 11
generator function, using GPT-4o 11-13
query, for user input 13-15
G
Galileo (spacecraft)
reference link 42
generative AI environment
installing 131, 132
generator 8, 124
augmented input with HF 8
content generation 132-134
generation and output (G4) 8
generative AI environment, installing 131, 132
HF-RAG for augmented document inputs, integrating 125, 126
input 8, 126
mean ranking simulation scenario 126
prompt engineering (G3) 8
Generator and Commentator 263, 273, 274
AI-generated video dataset 263
frames, commenting on 275-277
Pipeline 1 controller 277
video, displaying 274
videos, spitting into frames 274, 275
GitHub 261
H
Hubble Space Telescope
reference link 41
Hugging Face 213, 214
reference link 213
hybrid adaptive RAG
building, in Python 120
generator 124
retriever 121
I
index-based RAG 62
architecture 62-64
index-based search 21, 24, 25, 62
augmented input 25
feature extraction 25
generation 26
versus vector-based search 64
International Space Station (ISS)
reference link 41
J
Juno (spacecraft)
reference link 41
K
Kaggle
reference link 148
Kepler space telescope
reference link 42
keyword index query engine 74, 85, 86
performance metric 87
knowledge-graph-based semantic search
graph, building from trees 185-187
RAG architecture, using for 182-185
knowledge graph index-based RAG 195-197
example metrics 202-203
functions, defining 200
generating 196, 197
graph, displaying 198, 199
interacting 199, 200
metrics calculation 204-206
metrics display 204-206
re-ranking 201
similarity score packages, installing 200
knowledge graphs 181
L
Large Language Model (LLM) 3
list index query engine 74, 83, 84
performance metric 84, 85
Llama
using, with dynamic RAG 227-230
LLM dataset
loading 93-95
LLM query engine
initializing 95
textual dataset, querying 96
user input, for multimodal modular RAG 95
M
machine learning (ML) 146, 215
Mars rover
reference link 41
mean ranking simulation scenario
human-expert feedback RAG 128-130
no human-expert feedback RAG 130, 131
no RAG 127
metadata
retrieving 188-192
metrics
analyzing, of training process and model 249-251
metrics, fine-tuned models
reference link 249
ML model, training 154
clustering evaluation 156-158
clustering implementation 156-158
data preparation and clustering 154-156
modular RAG 4, 26-28
strategies 28
multimodal dataset structure
bounding boxes, adding 100-103
image, displaying 100
image, saving 100-104
image, selecting 99
navigating 99
multimodal modular RAG 90-92
building, for drone technology 93
performance metric 110
user input 95
multimodal modular RAG, performance metric 110
LLM 110
multimodal 111-113
overall performance 113
multimodal modular RAG program, for drone technology
building 93, 108-110
LLM dataset, loading 93-95
multimodal dataset, loading 96-99
multimodal dataset structure, navigating 99
multimodal dataset, visualizing 96-99
multimodal query engine, building 104
performance metric 110
multimodal query engine
building 104
creating 104, 105
query, running on VisDrone multimodal dataset 106
response, processing 106, 107
source code image, selecting 107, 108
vector index, creating 104, 105
N
naïve RAG 4
augmented input 20
example, creating with 18
generation 20, 21
keyword search and matching 18, 19
metrics 19, 20
O
ONNX
reference link 215
OpenAI 261, 262
URL 39
OpenAI model
fine-tunes, monitoring 244-246
fine-tuning 242, 243
for embedding 159
for generation 159
Pinecone constraints 159
Open Neural Network Exchange (ONNX) 215
P
Pinecone 262, 263
reference link 172
used, for scaling 144
Pinecone index
querying 282-286
Pinecone index (vector store)
challenges 159, 160
creating 166, 167
data, duplicating 165, 166
dataset, chunking 162
dataset, embedding 163-165
dataset, processing 161, 162
environment, installing 160
querying 170-172
scaling 158
upserting 168-170
Pipeline 1 controller 277-279
comments, saving 279, 280
files, deleting 280
Python
used, for building hybrid adaptive RAG 120
R
RAG architecture
for video production 256-258
RAG ecosystem 237, 238
domains 5-7
evaluator component 8
generator component 8
retriever component 7
trainer component 9
RAG framework
advanced RAG 4
generator 4
modular RAG 4
naive RAG 4
retriever 4
RAG generative AI 172
augmented generation 176-178
augmented prompt 176
relevant texts, extracting 175
using, with GPT-4o 172
RAG pipeline 33, 34
augmented generation 35, 50, 51
building, steps 36
components 34
data collection 35, 40-42
data embedding and storage 35, 44, 45
data preparation 35, 40-44
environment setup 36
reasons, for component approach 34
RAG, with GPT-4o 172, 173
dataset, querying 173
target vector, querying 173, 175
Retrieval Augmented Generation (RAG) 1-3, 50
non-parametric 4
parametric 4
versus fine-tuning 4, 5
retrieval metrics 15
cosine similarity 15, 16
enhanced similarity 17
retriever 121
data, processing 122, 123
dataset, preparing 121
environment, installing 121
user input process 123, 124
retriever component
collect 7
process 7
retrieval query 8
storage 7
S
scaling, with Pinecone 144
architecture 144-146
semantic index-based RAG program
building 64, 65
cosine similarity metric 75, 76
Deep Lake vector store, creating 69-74
Deep Lake vector store, populating 69-74
documents collection 65-69
documents preparation 65-69
environment, installing 65
implementing 74
query parameters 75
user input 75
session time
activating 215, 216
Silhouette score 156
space exploration
reference link 41
SpaceX
reference link 41
T
Term Frequency-Inverse Document Frequency (TF-IDF) 15, 57, 134
trainer 9
training loss 251
tree index query engine 74, 80-82
performance metric 83
U
upserting process
reference link 168
user interface (UI) 124
V
vector-based search
versus index-based search 64
vector search 21
augmented input 23
generation 23
metrics 22, 23
vector similarity
reference link 167
Vector Store Administrator 281, 282
Pinecone index, querying 282-286
vector store index query engine 74-76
optimized chunking 79
performance metric 79, 80
query response and source 77, 78
vector stores 33
Video Expert 286-291
video production ecosystem, environment 259
GitHub 261
modules and libraries, importing 260, 261
OpenAI 261
Pinecone 262, 263
Voyager program
reference link 42
W
Wikipedia data
retrieving 188-192
