Funddo
Automated analysis of financial documents and generate the results in natural language using AI models.
Customer has hundreds of financial documents and they were looking for a solution which can read & analyze the documents and provide the response to user’s query in natural language.
Client:
Trilasoft
Category:
Marketing/Finance
Funddo – a mobile lending platform of Areeva Products & Services Private Limited (APSPL) Information technology (IT) software development and product & services Company.
They are committed to use latest technology to promote and deliver digital financial products and services globally, and collaborating with the partners to provide sustainable, inclusive financial services to consumers, small and micro enterprises.
01. Problem Statement
The customer wanted to build and scale generative AI applications for efficient financial data analysis based on bank statements uploaded by users in PDF format.
They want automate the process of creating a dataset from the uploaded bank statement files and building a comprehensive financial model and ask any questions related to bank statement.
Below were the requirements –
- The user uploads the bank statement in pdf format
- Creating a dataset of the uploaded BS file
- Building LLM
- Present Analysis Data
02. Proposed Solution & Architecture
Following steps are performed to setup multi user access of the system and block data transfer –
- Source data was uploaded to AWS S3 which could be bank statements or any other PDF files
- Amazon Textract tool was be used to extract data from PDF and convert to textual format which was stored in Output S3 bucket
- Amazon Athena was used to query data from target S3 bucket and feed the data in tabular format through query
- Bedrock will use Athena table as data source and Jurassic-2 Ultra as LLM model for text generation
- Athena was integrated with QuickSight for data visualization, dashboards and reports.
our industries
Service Used In Funddo
We help you see the world differently, discover opportunities you may never have imagined and achieve results that bridge what is with what can be.
AWS CloudTrail
AWS Config
Amazon GuardDuty
Amazon CloudWatch
Amazon EC2
Amazon Textract
Amazon Athena
Amazon S3
QuickSight
BedRock
03. Results
- Successful deployment of AWS infrastructure
- Successful response to user queries in natural language from the application using Bank Statement PDF as data source
- Successful creation of Dashboards and reports in AWS QuickSight to showcase financial metrics and its value
04. TCO Analysis Performed
- 1 Year TCO – $36K
- 3 Year TCO – $120K
