[Day 137] AWS Summit Seoul Day 2

 Hello :)
Today is Day 137!


A quick summary of today:
  • attended 'data analysis' and 'business tech' sessions during Day 2 of AWS Summit in Seoul


I tried to take more notes with my laptop during the session today.


First session - 'Data strategy for successful GenAI on AWS'

A GenAI app is like an iceberg. 

Under the water it consists of among other things data storage, database, data lake, data manipulation, data governance. 

The session introduced us to RAG, and how we can either use an out-of-the-box language model, pre-train a model, or train a model from scratch. 

As for a vector search AWS offers plenty of options: Amazon OpenSearch Service, OpenSearch Serverless, Aurora PostgreSQL, RDS for PostgreSQL, DocumentDB, DynamoDB via zero ETL, MemoryDB for redis, Neptune.

As for a database, there are plenty of structured and nonstructured dbs. 

Amazon Bedrock - allows for using out-of-the-box models for GenAI applications. It gives us the ability to use a simple API for a model, allows for model customizing, RAG, Agents, protection, privacy and security. 

If we use Amazon Bedrock, the vector databases we can use are Amazon OpenSearch Serverless, Redis Enterprise Cloud, Pinecone, Amazon Aurora, MongoDB.


Second session - 'How to best use vector databases on AWS'

Start with a pdf - chunk - embed with Amazon Tital - store in Amazon Aurora PostgreSQL (or other db)

Things we need to look for when embedding. 

  • embedding time
  • size
  • query time

Questions to ask when choosing a db system

  • for my workflow, which vector storage is the best?
  • how much data am I going to store?
  • among storage, efficiency, connectiviy, cost which is important to me?
  • what are the pros and cons of the different indexing, query time and schema designs?
What is pgvector?
  • offers open source similarity search, vector storage, indexing and metadata
From pgvector's github it has 2 strategies for indexes: 

HNSW: An HNSW index creates a multilayer graph. It has better query performance than IVFFlat (in terms of speed-recall tradeoff), but has slower build times and uses more memory. Also, an index can be created without any data in the table since there isn’t a training step like IVFFlat.

IVFFlat: An IVFFlat index divides vectors into lists, and then searches a subset of those lists that are closest to the query vector. It has faster build times and uses less memory than HNSW, but has lower query performance (in terms of speed-recall tradeoff).


Third session - 'Successful data stories from the business'

Presented many cases (companies) where after integrating snowflake into their operations, the companies reduced cost, increased customer satisfaction, profits, amount of data, customer service. Cases like KFC, AT&T, Pfizer, SCANIA.


Below is what kind of services are offered


Basicaully, it was a sales pitch for snowflake haha


Fourth session - Improving GenAI search performance through Amazon's OpenSearch

What is Amazon OpenSearch? The most optimized search and log solution
  • offers similarity search
  • seamless data analysis and visualisation
  • connects data sources
  • low cost solution
  • easy of use
Introduced vector search and TF-IDF scoring, and the process of giving a PDF, chunking, embedding, saving, finding similar to a query passages and retrieving them.

Using the ANN algorithm - Hierarchical navigable small world (HNSW), and IVF

Showed an example of semntic search:

  1. external model hosting and service connection
  2. creating an OpenSearch document gathering using NN search pipeline
  3. search request through the API gateway
  4. call the backend through the API gateway
  5. search the backend for related documents and after give them to the client

We saw a demo on using Amazon OpenSearch and Bedrock for text2sql - similar to my chat with your PDFs but of course on a much higher level using OpenSearch and using streamlit. Also every query, response and extra metadata gets saved in the backend.

Below is what happens on a higher level


Fifth session (from the business tech sessions because there is no data analysis session in this time slot) - Easy data lake creation using AWS

What is a data lake? A place that stores all kinds of data (structured and unstructured)

Data lake vs Data warehous

Data saved in any format (raw) vs fixed format

Schema is not needed vs Schema is needed

Low trustability vs High trustability

Storage and volume cost are better vs Provides fast and efficient queries


Data lake key features (example - Amazon S3)

  • data catalogue and searching
  • user access
  • data gathering
  • manipulation and analysis
There are also data lake houses that combine the benefits of both lakes and houses

Things to consider for data lakes in finance

  • regulation compliance
  • multi-user support
  • user certification and ability to use/maintain
  • authorization
  • protection against attackers
  • private network connection and traffic
Data lake security levels
  • identity and access management
  • application security
  • data protection
  • infrastructure security
  • network security
  • identity and access management (platform)
  • multi-account management
AWS lake formation


Sixth session - 'Using Amazing SageMaker Canvas for GenAI'

Data prep challenges

  • too many different tools
  • various code dependancies
  • scalability
  • operations management

Data prep/manipulation to get a clean version of the data for a model can take 60-80% of the time of an ML project

SageMaker Canvas has us covered - new feature - chat for data prep

  • run code with plain text, edit/remove/add data, get insights, data prep process one with no code and ready to make a model

After our data is ready, there are ready to use models for text, image, audio content. 

Finetuning? Can be done without writing any code as well. Selecting a model (or up to 3), data, and we get training and validation results and extra info if we want to finetune with python.


Great day. I am glad I attended ^^


That is all for today!

See you tomorrow :)

Popular posts from this blog

[Day 198] Transactions Data Streaming Pipeline Porject [v1 completed]

Image
 Hello :) Today is Day 198! A quick summary of today: data streaming pipeline project [v1 done] Here is a link to the project's repo. Well ... I did not know I can do it in a day (~14 hours) after yesterday's issues but here we are. Turns out in order to insert the full (~70 variables with nested/list structure), I need the proper pyspark schema. And yesterday I did not have that and that is why when I was reading data in the kafka producer I was getting NULL in the columns - my schema was wrong. Well today I not only fixed the schema for the 4 variables I had yesterday, but included *all* the variables that come from the Stripe API ~ 70 (for completeness).  When I run docker-compose, the data streams and is input into the postgres db (and is still running). Unfortunately, the free Stripe API for creating realistic transactions has a limit of 25, so every 3 seconds, 25 new transactions are sent to the db. It has been running half the day (since I got that set up) and as I am wr
Read more

[Day 107] Transforming natural language to charts

Image
 Hello :) Today is Day 107! A quick summary of today: created text2chart - transforming natural language to charts [ github ] [ webapp ] I am super sick, but in the around 3 hours of my fever not being high, I managed to study a bit. I was again thinking about creating some nlp project but was stuck on the idea of how to host a model. After looking around I found this article where they create a similar app called chat2vis  and they use codellama. I know codellama is an open source model that I can get from huggingface and I wondered, how did they host it. Turns out - they ask the user to input a huggingface api key, which then allows for querying and generating content. (AWESOME! this is applicable to some of the other PDF chat apps that I have, but I will look to make an app of them later).  My fever eased a bit, and I sat on my chair to do some work ^^ Here is a breakdown of the app. I used streamlit. Also, 2 default datasets during dev: Financial Statements of Major Companies(2009-
Read more

[Day 54] I became a backprop ninja! (woohoo)

Image
 Hello :) Today is Day 54! A quick summary of today: Read through cs231n's neural network lectures [1] [2] [3] Spent 8 hours going over Andrej Karpathy's backprop ninja tutorial After yesterday's confidence boost in doing backprop, I came in today with the intention of becoming of, as Andrej Karpathy calls it, a backprop ninja.  The cs231n NN lectures were the usual about how NN became to be, param initialization, activation functions, loss functions, regularization and optimizers. Nothing out of the ordinary, but still was good to reassure myself of that knowledge. But that was definitely not the main even of today. As I mentioned, I wanted to become (or at least attempt to become) a backprop ninja today. I have watched that particular tutorial twice, the 1st time I had no idea what is going on, the 2nd time I could follow, and now it was the 3rd time - and now, I wanted to participate and actually see the why and the how of the things we have to do, when we do manual back
Read more