Streamlined RFI Response Workflow for a Fintech Company

Our innovative web application streamlines the process of responding to Requests for Information (RFIs) by leveraging a Knowledge-based Question Answering (QA) System. Utilizing the wealth of knowledge contained within previous RFIs, our system embeds this information into a vector database. When a new RFI is submitted, our QA system retrieves similar QA pairs from the knowledge base, generating accurate and relevant answers efficiently.

Key features include:

• Automatically fills RFIs with optimal answers rapidly.
• Seamless integration of Large Language Models (LLMs) for natural language understanding and generation.
• Utilizes RAG methodology for generating comprehensive and context-rich responses.
• Implements various RAG approaches including Naïve RAG, Window Sentence RAG, and Hierarchical RAG to cater to specific data requirements.
• The system was deployed on AWS EC2 servers

Results:

• Reduces RFI screening and filling time from months to a fraction, ensuring timely project execution and adherence to timelines.
• Minimizes errors in information extraction and response generation, thereby improving project quality and client satisfaction while reducing operational costs and boosting productivity.

Tech-Stack:

Python             OpenAI            Chroma-db      FastAPI           React.js           AWS EC2

Background:

Netsol, a leader in asset finance software for car financing and leasing, faced a challenge. Their clients, often lacking strong SQL expertise, relied on technical teams to generate reports and data queries. This created bottlenecks and slowed down access to crucial business insights. Netsol sought a solution to empower their everyday users to independently retrieve data using natural language, eliminating the need for complex SQL queries.

Our Solution:

We developed a custom Text-to-SQL framework seamlessly integrated with Netsol's legacy ASCENT system. This user-friendly interface allows non-technical users to formulate data queries in plain English. The system translates these queries into optimized SQL, retrieves the relevant information from the ASCENT database, and presents the results in a clear and concise format.

Key Benefits:

• Increased User Autonomy: Non-technical users can now fetch data independently, reducing reliance on IT teams and speeding up workflows.
• Improved Efficiency: Streamlined data retrieval eliminates the need for writing complex SQL queries, saving time and resources.
• Enhanced Business Insights: Faster access to data empowers users to make informed decisions based on real-time information.
• Simplified Reporting: Users can easily generate reports without technical barriers, improving operational transparency.
• Reduced Training Burden: The intuitive interface eliminates the need for extensive SQL training for everyday users.
• Seamless Integration: The Text-to-SQL framework seamlessly integrates with existing ASCENT infrastructure, minimizing disruption and maximizing ROI.

Technical Details:

• Tokenize and Encode text query and database schema items such as table and column names, and their foreign-key relationships using an off-the-shelf T5 tokenizer and a pretrained RoBERTa model.
• Train a Schema Item Classifier for the selection of relevant tables and columns
• Generate the SQL Query skeleton and fill in the column and table names using a T5 decoder.

Outcome:

The Text-to-SQL framework empowered Netsol's users to unlock the power of their data independently. This resulted in increased efficiency, improved access to business insights, and a more empowered user base. This successful project demonstrates our expertise in developing innovative AI/ML solutions that bridge the gap between technical capabilities and real-world business needs.

Welcome to the Enterprise Chatbot designed specifically for ITU University. We've meticulously gathered and curated data pertaining to faculty, scholarships, and admissions at ITU University. Our comprehensive pipeline, meticulously crafted after thorough data cleaning, integrates cutting-edge technologies to ensure optimal performance.

Our pipeline boasts a robust architecture, featuring a question classifier, retrieval-augmented generation (RAG) module, code generation module, and the powerful GPT-3.5-Turbo-1106 as our Large Language Model (LLM). This sophisticated setup enables us to generate responses that closely mimic human conversation, enhancing user experience significantly.

To complement our advanced backend, we've developed an intuitive frontend using React, seamlessly connected to our backend pipeline via Flask. This integration ensures a smooth user experience while harnessing the full potential of our chatbot.

Deployment of our model is handled efficiently through AWS EC2 and S3 bucket, ensuring reliability and scalability. It's worth noting that our chatbot is tailored exclusively for ITU University, providing accurate responses to queries specific to its domain.

Key technical features include:

• Seamless integration of Large Language Models (GPT-3.5-Turbo-1106) for natural language understanding and generation.
• Utilization of Chroma DB of Langchain to store vector embeddings created by the Ada-V2 embedding model.
• LLM-based classifier and code-generation module.
• Effective configuration of user queries before employing RAG, utilizing techniques such as generating multiple similar user queries from a single input, converting queries into SQL, and employing the HyDE method to convert queries into general documents.
• A blend of RAG techniques including Naïve RAG, Window sentence RAG, and Hierarchical RAG, tailored to our specific data requirements.
• Utilization of Flask web interface to seamlessly communicate with the frontend developed in React.
• Deployment using AWS EC2 and S3 bucket for both frontend and backend for demonstration purposes.

Results:

• Enhanced Student Support: Experience significantly improved query responses tailored to students' needs.
• Personalized and Accurate Assistance: Receive personalized and knowledge-based responses surpassing typical rule-based chatbots.
• Seamless 24/7 Support: Access perfect 24/7 customer care, ensuring continuous assistance and support.

Technology Stack:

Python           OpenAI          Flask              React.js          AWS

Demo Link:

This demo is specifically tailored for ITU University, offering a sophisticated chatbot solution designed to efficiently address user queries pertaining to faculty, admissions, and scholarships at the institution.

Please click on the following link to test our Enterprise Chatbot.

http://enterprise-chatbot.s3-website-us-east-1.amazonaws.com

Challenge: Navigating the complexities of immigration applications can be overwhelming for individuals and families. Lengthy forms, confusing legalities, and the potential for costly mistakes create a significant barrier to entry. Many applicants lack the resources to hire lawyers, leaving them feeling lost and unsure.

Solution: We developed an innovative solution utilizing Large Language Models (LLMs) and Optical Character Recognition (OCR) to streamline the immigration form-filling process. This solution takes the form of an interactive chatbot assistant, designed to:

• Conversational Guidance: The LLM-powered chatbot interacts with users in a natural, conversational manner. It guides them through the immigration application process, asking clear and concise questions to gather necessary information. The chatbot adapts its questioning based on user responses, ensuring a personalized experience.
• Intelligent Form Completion: As the user interacts with the chatbot, the LLM automatically populates the corresponding sections of the application for Asylum
• and for Withholding of Removal with the gathered information. This eliminates the need for manual form filling and reduces the risk of errors.
• OCR Integration: The solution integrates OCR technology, allowing users to upload scanned documents like passports, government IDs, and educational degrees. The OCR system extracts relevant information (e.g., names, dates of birth, educational credentials) and automatically populates it into the appropriate form fields.
• Legal Knowledge Base: The LLM leverages a vast legal knowledge base specific to immigration laws and procedures. This allows the chatbot to offer guidance and answer user questions related to eligibility requirements, specific application types, and potential complexities.

Benefits for Users:

• Increased Accessibility: The user-friendly chatbot interface empowers individuals and families to navigate the immigration process independently, reducing reliance on expensive lawyers.
• Enhanced Accuracy: AI-powered form completion minimizes errors and inconsistencies often associated with manual form filling.
• Reduced Time Commitment: The chatbot streamlines the process, significantly reducing the time it takes to complete immigration applications.
• Improved Clarity: Interactive guidance and access to a legal knowledge base provide users with a better understanding of the application process and their rights.

Impact and Future:

The AI-powered chatbot solution provides a valuable tool for simplifying immigration procedures. Early results show a significant increase in user application completion rates and a decrease in errors. This case study demonstrates the potential of AI to democratize access to legal processes and empower individuals to navigate complex systems with greater confidence and efficiency. As we continue to develop and refine the solution, we envision expanding its functionalities to support different immigration pathways and languages, further advancing accessibility and inclusivity in the immigration process.

Challenge: Stratton Capital, a venture capital firm, faced a growing workload as they evaluated a high volume of potential investment opportunities. Manual contract review, a crucial step in due diligence, was a time-consuming and resource-intensive process. This slowed down their investment decision-making and limited their capacity to explore new deals.

Solution: We developed an AI-powered contract analysis solution to streamline Stratton Capital's due diligence process. This solution leveraged a combination of cutting-edge technologies:

• Deep Learning for Contract Classification: We utilized a pre-trained deep learning model specifically designed for legal document classification. This model, fine-tuned on a vast dataset of various contract types, could automatically categorize incoming contracts (e.g., NDAs, employment agreements, service agreements).
• Named Entity Recognition (NER) for Key Clauses: The solution employed NER techniques to identify and extract critical information from within contracts, such as termination clauses, non-compete agreements, and intellectual property (IP) rights.
• Risk Assessment and Flagging: By analyzing the extracted information and applying pre-defined risk rules, the solution could highlight potential risks or red flags within the contracts, enabling Stratton Capital to prioritize issues requiring further attention.

Deployment and User Interface: The solution was seamlessly integrated with Stratton Capital's existing document management system, allowing for a smooth workflow. Users could upload contracts for analysis directly within the system, and the AI would generate a clear and concise report summarizing key clauses, potential risks, and relevant sections requiring further review.

Key Benefits:

• Increased Efficiency: Automating contract classification and key clause extraction significantly reduced review time for Stratton Capital's investment team.
• Enhanced Accuracy: The AI model's ability to identify and categorize contracts, combined with NER for clause extraction, ensured consistent and accurate analysis, minimizing human error.
• Improved Risk Management: Automated risk assessment and flagging highlighted potential issues early on, allowing Stratton Capital to make informed investment decisions.
• Focus on Strategic Issues: By freeing up time from manual review tasks, the investment team could dedicate more resources to analyzing deal terms and conducting deeper due diligence.
• Scalability and Adaptability: The cloud-based deployment and modular design of the solution allowed for easy scalability as Stratton Capital's workload grew and for adaptation to accommodate new contract types or risk factors in the future.

Outcome: Stratton Capital experienced a dramatic improvement in their due diligence process efficiency. Contract review time decreased by an estimated 70%, allowing the investment team to evaluate a higher volume of deals and make quicker investment decisions. The AI-powered solution also improved risk management and ensured consistent analysis across all contracts. This case study highlights the potential of AI to automate repetitive tasks in the legal industry, empowering investment firms like Stratton Capital to operate with greater speed, accuracy, and efficiency.

Streamlined RFI Response Workflow for a Fintech Company

Our innovative web application streamlines the process of responding to Requests for Information (RFIs) by leveraging a Knowledge-based Question Answering (QA) System. Utilizing the wealth of knowledge contained within previous RFIs, our system embeds this information into a vector database. When a new RFI is submitted, our QA system retrieves similar QA pairs from the knowledge base, generating accurate and relevant answers efficiently.

Key features include:

• Automatically fills RFIs with optimal answers rapidly.
• Seamless integration of Large Language Models (LLMs) for natural language understanding and generation.
• Utilizes RAG methodology for generating comprehensive and context-rich responses.
• Implements various RAG approaches including Naïve RAG, Window Sentence RAG, and Hierarchical RAG to cater to specific data requirements.
• The system was deployed on AWS EC2 servers

Results:

• Reduces RFI screening and filling time from months to a fraction, ensuring timely project execution and adherence to timelines.
• Minimizes errors in information extraction and response generation, thereby improving project quality and client satisfaction while reducing operational costs and boosting productivity.

Tech-Stack:

Python             OpenAI            Chroma-db      FastAPI           React.js           AWS EC2