Chat with Financial Reports

An international private equity firm approached us after identifying a key pain point experienced across its investment team. Reviews of financial reports (e.g., HSBC) and portfolio companies' quarterly reports take 45 minutes - 1 hour per person per report each time a new release is issued.

Managers and analysts search for insights on specific geographies, growth rates, and inflation. They compare new releases with previous ones to identify investment trends. Searching for a specific facts can take as much as 10-15 minutes. The process is not only time-consuming, but it can lead to missing key insights buried within the pages.

The solution was clear for them: leverage Generative AI to assist with search. They partnered with Stack AI for our enterprise-grade environment, ease of use, short time to production, and cost savings (~80% less expensive than in-house development).

In less than 3 weeks, they deployed an AI assistant in their VPC on Azure, controlling their confidential data end-to-end.

With AI, they reduced the search time by ~67% and enhanced the quality of their investment decisions.

Follow this guide to deploy a similar AI assistant for your organization.

Creating an AI Assistant

Available models in Stack AI

First, select the right model for your use case and company context. Our customer chose an open-source model to host in their VPC. After testing various models in Stack AI (Llama 3, Mixtral, Google’s Gemma, etc.), they opted for a Mixtral Large. In Stack AI, you don’t need to set up an account with any provider. We handle it for you, allowing easy experimentation.

Add instructions in the prompt to guide the Large Language Model on how to answer user questions. Follow this prompting guide to optimize your instructions.

Financial Assistant Workflow

Set up the knowledge base used by the AI model to extract facts. Stack AI integrates with most common database providers (e.g., Sharepoint, OneDrive, AWS S3), enabling secure connections with your applications. Select the folders where the information is stored and start interacting with your chatbot.

This approach works out of the box. However, if you want to customize the data pipeline to your specific data structure (i.e., specific sections length), Stack AI offers the following.

Chunk Visualizer

With the chunk visualizer, you will see how information from a selected document is decomposed into chunks of text, which are then sent to the LLM for context. Different chunks are sent depending on the user's query, enhancing the efficiency of the LLM - known as RAG or Retrieval Augmented Generation.

Enable features to improve information retrieval such as query transformation in the settings of the knowledge base. If you want to further deep-dive into these advanced features contact us!

Exporting the AI Assistant

Export Tab

Stack AI offers pre-built interfaces for exporting your application. The most common one is the chatGPT-style interface where users ask questions and create threads.

Other options include the form interface, ideal for guiding users with the inputs required to generate an output or multiple outputs. Batch processing is ideal for running workflows over a list of scenarios provided via CSV or manually. Deployment options also include Slack, WhatsApp, and SMS. We recommend you this guide on how to deploy a Slack bot. We also provide API endpoints to build your own user interface.

Citations help understand where the AI model sources information. Read the source file and specific chunks used for responses by clicking a citation. Enable this feature in your LLM node settings.

In the Export tab, you'll find various customization options (name, logo, colors) to match your brand's look and feel. Configure a custom domain if needed and protect your chatbot with SSO or password.

Monitoring

Analytics Tab

Monitor usage and application performance in the Analytics tab. Generate reports with insights automatically through our AI Report Writer. Dive deep into users’ conversations with the AI assistant to enhance the chatbot's logic.

Security

We offer on-premise options for your applications, either in your VPC or on your own servers. If you choose our cloud solution, rest assured we are committed to data protection and security, being SOC 2 Type II, GDPR, and HIPAA compliant. We partner with security leaders such as Secureframe and Modern Assurance to ensure enterprise-grade security.

What’s Next?

Using Stack AI reduces the cost of AI assistants deployment by ~80% compared to in-house development, allowing organizations to focus on primary business challenges without the burden of technological development.

Considering an AI assistants for knowledge management? Schedule a meeting with us to explore our solutions.