Custom ChatGPT-Style AI for Investment Firm

BUSINESS REQUIREMENTS

The investment advisory firm aimed to harness AI to enhance its client interaction processes and deliver personalized investment advice. The firm envisioned a ChatGPT-style AI investment advisor capable of understanding intricate financial terminologies, answering customer inquiries accurately, and adapting to shifting investment trends. Importantly, this AI advisor also needed to adhere to stringent data privacy regulations, a characteristic requirement in the financial sector.

CHALLENGES

The project encountered several key challenges:

• Domain-specific Language Understanding: The ChatGPT-style AI needed to comprehend and generate responses in line with intricate financial and investment terminologies.
• Data Security and Privacy: With financial data being highly sensitive, the AI system needed to adhere to stringent data protection and privacy regulations.
• Scalability: The solution should be capable of handling a high volume of user interactions, with potential for scalability as the business grows.
• Commercial Usability: The system needed to be both open-source (for flexibility and customization) and commercially usable (for long-term sustainability). Most LLM models that are Open-Source cannot be used commercially.

SOLUTIONS

To overcome these challenges, we utilized Dolly2 and Langchain to create a custom, domain-specific ChatGPT-style AI model.

• Data Collection and Preparation: The firm provided comprehensive datasets, including financial literature and client interaction transcripts. These datasets were cleaned and preprocessed to remove any personally identifiable information (PII) and formatted to align with Dolly2’s training requirements.
• Domain-specific Training: The cleaned data was used to fine-tune Dolly2 using Langchain’s distributed training architecture. This was done on the firm’s secure servers to ensure data privacy. The fine-tuning process involved training the ChatGPT-style AI to understand and generate responses in line with intricate financial terminologies and investment strategies.
• AI Model Deployment: The trained ChatGPT-style AI model was deployed on-premise using Langchain’s efficient architecture. This allowed the model to handle a high volume of user interactions while ensuring that all data processing occurred within the firm’s secure network.
• BPM with Camunda: The firm utilized Camunda as their BPM system to orchestrate the entire process. Camunda was used to manage the workflows and decision-making processes, integrating the AI model with the firm’s client interaction platform. This ensured a seamless and efficient operation, from receiving a client inquiry to providing an appropriate response.
• User Interface Integration: The deployed ChatGPT-style AI model was then integrated into the firm’s client interaction platform. This allowed the AI model to handle client inquiries directly.
• Monitoring and Updates: The system was set up to allow for continuous monitoring and updates. This ensured that the ChatGPT-style AI model could adapt to changing financial trends and terminologies.
• Commercial Usability: The open-source nature of Dolly2 allowed for the required flexibility and customization. Importantly, unlike many other open-source models, Dolly2 is commercially usable, ensuring the sustainability and scalability of the solution.

IMPACT

The custom ChatGPT-style AI model has significantly transformed the firm’s client interaction processes:

• Improved Customer Experience: The ChatGPT-style AI investment advisor could answer intricate financial inquiries accurately, leading to enhanced customer satisfaction and trust.
• Enhanced Efficiency: The ChatGPT-style AI system handled routine inquiries, freeing up the firm’s personnel to concentrate on complex tasks and strategic decision-making.
• Scalable Client Interaction: The ChatGPT-style AI system allowed the firm to interact with a large number of clients simultaneously, a feat that wasn’t possible previously.
• Data Privacy Compliance: The on-premise training and inference ensured that all data privacy regulations were adhered to, and the clients’ financial information remained secure within the firm’s network.

KEY TECHNOLOGIES

• Camunda
• Dolly2 Transformer Model
• Langchain
• Java
• HuggingFace