Deploy to Secured User Access

1. Kalika Sales Dashboard visualization

2. Kalika Dify Agent 3, Kalika Sales and Marketing agent

3. Prompt Workflow

4. AWS Services

5. GoogleCloud and Agentspace

RAG

Basic RAG Architecture

RAG Usecases

Gmail RAG

1.Data Extraction: Extract the PO order and proforma invoice with attached file by searching keyword to S3

starting point

Problem Statement : Build RAG on Kalika Enterprises Gmail Data

Tasks

1. Understand the details of various documents i) Proforma Invoice:A proforma invoice is a preliminary bill that a seller sends to a buyer before a sale is confirmed. It's a non-binding document that's used for planning, budgeting, and estimates. It's also used for customs clearance and financing ii) PO Dump : Pending and Processed data details

2. Create the Parsing process

   1. Proforma Invoice

      • Searching a with proforma invoice for smtp mail python utility download attached file
      • Dump that on scheduling daily to s3
      • Extract the content using ocr engine and validate try:
        1. Python package
        2. Medium article

   2. PO dump _ On Daily basis Dump is in mail extract excel store in s3

      • Implement pg vector

   TODO

   1) Create PG vector on local
   2) Upload po_dump excel in PG vector and create a vector
   3) Test with Ollama local model and create a streamlit app
   4) Test for last 10 days documents with query and response
   5) Generate the report and share it with in team group and github
   

3. Deploy on Ec2 for further building POC

Financial Assistant RAG : B2C

Problem Statement : Uploading Bank statement, Trading Documents(Last trading transaction,orders session details)

give query response bot.

1. User can upload documents such as bank statement, trading sheets with Upload button- UI
2. create PG vector to store
3. Ollama testing with local model

Sample Datasets

Kaggle

Link to Refer

medium

Limitation : Data privacy

Solution: Give a isolated space

Points of discussion

1. Explore the techniques for giving data privacy to user

Various Vector Database

Here are some notable open-source frameworks and libraries for building Retrieval Augmented Generation (RAG) systems:

1. SWIRL: An infrastructure software that facilitates secure and fast searches across data sources without the need for data movement. It integrates with over 20 large language models (LLMs) and is designed for secure deployment within private clouds.

2. Cognita: A modular framework that organizes RAG components for easier testing and deployment. It supports various document retrievers and is fully API-driven, making it suitable for scalable RAG systems.

3. LLM-Ware: This framework focuses on enterprise-ready RAG pipelines, allowing the integration of small, specialized models. It supports a modular architecture and can operate without a GPU.

4. RAG Flow: An engine that emphasizes deep document understanding, enabling effective integration of structured and unstructured data for citation-grounded question-answering.

5. Graph RAG: A graph-based system that enhances LLM outputs by incorporating structured knowledge graphs, making it ideal for complex enterprise applications.

6. Haystack: An orchestration framework that connects models, vector databases, and file converters to create advanced RAG systems, supporting customizable pipelines for various tasks.

7. Storm: A knowledge curation system that generates comprehensive reports with citations, integrating advanced retrieval methods to support multi-perspective question-asking.