The purpose of hackathon is to leverage AI and ML Technologies to address critical challenges in the financial markets. The overall goal is to progress industry through Innovation, Networking and by providing effective Solutions.
Hackathon Key Dates • June 6th - Event invites will be sent to participants • June 9th - Hackathon Open • June 9th-11th - Team collaboration and Use Case development • June 12th - Team presentations & demos • June 16th - Winners Announcement
More Info - https://communications.dtcc.com/dtcc-ai-hackathon-registration-17810.html
Commit Early & Commit Often!!!
Tax Event Detection & Categorization
The project is accessible for the moment at the url http://34.219.206.185:8501/
Our solution is a AI-driven platform that users can rely on for their tax-related issues. The platform has two major components – a chat system where users can ask questions about financial matters to get reliable answers with citations, and an alerting system where users can subscribe to financial-tax topics relevant to them.
The core of the system works with a knowledge base composed of credible, reliable sources such as officially released government circulars and top news providers. New information is periodically fetched/scraped from pre-decided online sources, which then:
- Becomes part of the knowledge base
- Gets categorized into one of the subscription categories, triggering an alert to the subscribers.
This helps our platform always stay updated with the most recent information.
The combination of web scraping, RSS feeds, and publicly available APIs is used to fetch data. These scrapers and fetchers are in the form of Lambda functions taxLLMScraperFunction and incometaxGovScraper that get triggered periodically, storing the information in an S3 bucket.
The documents stored in the S3 bucket are run through the Named Entity Recognition service provided by AWS Comprehend to identify the relevant entities referenced in a document. This information is made part of the metadata to ensure better knowledge retrieval and prioritization.
Whenever a document is added to S3, it triggers a Lambda function classifyTaxArticle that uses an LLM model to categorize the new information into one of many pre-determined categories mentioned below.
- Tax Policy or Regulatory Updates
- Deadline Changes
- Global Tax Updates
- Litigations and Rulings
- Technology and Compliance Innovations
- Funding & Mergers and Acquisitions
- Security and Fraud Alerts
- Emerging Tech Trends
- General Updates
This classification is stored in DynamoDB ArticlesClassification for later access, which in turn triggers another Lambda function sendEmailNotifications to send out emails to subscribed users.
Amazon Kendra acts as our main search engine. On a periodic basis, Amazon Kendra goes through the S3 buckets containing the curated documents and their metadata, indexing the information for easy retrieval. Kendra, in addition, crawls through a number of relevant websites, adding more information to the knowledge base system.
The chatbot part of the application utilizes taxLLMRAGFunction as the interface to this application. This Lambda is responsible for accepting user queries, fetching relevant context snippets from Kendra, and then querying the Bedrock LLM model to generate reliable answers back to the users.
- In AWS Bedrock, guardrails are used to redact any PII from both the user query as well as the LLM response.
- An on-prem server-hosted ML NRE model, which will also redact PII from the user query can be used. The reasoning behind this approach is to have the benefit of making user query anonymous even before it leaves the company network, making it more secure.
To simulate this, the set up is done on an EC2 server powered by the open-source model –iiiorg/piiranha-v1-detect-personal-information.
- User authentication system
- Chat history
- News feed of subscribed topics
- Alerting system based on categorized tax updates
- Convert any numerical figure in the prompt into a range to support redaction or normalization.
- Integrate a quantitative data source and processing engine to support mathematical and tax-related calculations.
- Monitor for any suspicious or illegal activities and flag the user.
- Nanda Kumar
- Sidharth Anil
- Anagha M
- Kavya Gonuguntla
All commits must be signed with a DCO signature to avoid being flagged by the DCO Bot. This means that your commit log message must contain a line that looks like the following one, with your actual name and email address:
Signed-off-by: John Doe <john.doe@example.com>
Adding the -s flag to your git commit will add that line automatically. You can also add it manually as part of your commit log message or add it afterwards with git commit --amend -s.
See CONTRIBUTING.md for more information
Copyright 2025 FINOS
Distributed under the Apache License, Version 2.0.
SPDX-License-Identifier: Apache-2.0